E-waste is a massive environmental problem. So are current data center plans, if recent reports are to be believed. However, a team of researchers from the University of California, San Diego, have come up with an intriguing idea: They plan to use 2,000 Google Pixel smartphones to build a cloud computing data center with already-existing tech.

According to a Google Research blog post, on average, people replace their smartphones every four years (via Hothardware). However, many modern (yet outdated, in terms of our constant desire for shiny new things) examples have processors, memory, and storage chips that are relatively powerful, particularly when you chain them together.

That's wasted hardware, and an ecological concern when you think of the extra carbon emissions created by manufacturing their replacements. By putting the existing chips to good use, it prevents them from going to landfill—and might even negate the need for new hardware in certain applications.

The post's authors say that the single-threaded performance of a modern smartphone's processor cores is on par with (or better than) many multicore server chips. However, modern servers are made up of dozens of multithreaded processor cores with access to a huge amount of memory, whereas a typical older smartphone only has a handful of cores and around 8-12 GB to play with.

Not only that, but recycled smartphones have a lot of extra components that would be inefficient (or hazardous) to deploy en masse, like the batteries and displays.

So, the first step is to remove everything but the motherboard and the attached chips (which represent the most embodied carbon of all the components), before chaining them together to create a server cluster for university usage, targeting relatively lightweight applications.

The phones are orchestrated together by Kubernetes, an open-source system for managing containerized applications. Each has a Linux distro installed, bypassing Android systems that wouldn't be suitable for mass-server deployment, like memory-saving features.

While the current iteration seems to be pretty small-scale, the eventual 2,000-phone data center is planned to be used for grading and research applications within the universities' existing software infrastructure.

"Early experiments show that even a moderately-sized cluster of 20 phones is capable of supporting peak submission rates for a 75+ student class, with grading latencies below the default AWS backend," say the researchers. "A 2,000 phone deployment will be capable of supporting a hundred such classes at once."

The post's authors say that Google will be supporting the project, and that the aim is to provide "hundreds of researchers and students with low-cost, low-carbon cloud computing, reducing the need for newly manufactured hardware and their associated emissions."

Which is such a neat idea, I'd like to see more officially supported projects like it in future. In my own home, I can think of at least five smartphones sitting in drawers doing absolutely nothing, all of which contain chips that could be used for something useful.

And while plugging them back in would of course lead to unused chips drawing power from the grid once more, I suppose it beats them being trampled by bulldozers at my local dump at some point in the future. Although it must be said, I doubt this work will do much to offset the ecological concerns around Google's own huge data center plans in the near future.

A ruling from a German court has found that Google is liable for the claims made in Search's AI Overviews. What is this? The consequence of Google's all-in-on-AI actions?

The case involves false claims made about two Munich-based publishers. Allegedly, Search's AI Overview misattributed the questionable practices of another existing business to the plaintiffs, drawing a link that did not exist in the sources it scraped. The two publishers initially sent a cease-and-desist letter to Google, only bringing the legal case after the search giant did not appropriately address the issue (via The Decoder).

As a result, on May 28, the Munich Regional Court issued an injunction against Google. To get a little bit into Deutschland's legal landscape, there are existing rulings from Germany's Federal Court of Justice that basically say companies like Google have limited liability when it comes to the third-party content dredged up by traditional search results. The Munich Regional Court argues that AI Overviews represent a different legal beast, and its ruling could have an international impact in the future.

The court makes the case that, from the perspective of your average user, the AI-generated response reads closer to direct information from Google rather than pointing towards external content (via Heise Online). Considering Pew Research found last year that Google users are much less likely to click on a source shared via an AI Overview, I can definitely follow the argument.

According to The Decoder's translation of the court documents, the court argued that Google owns the content its AI Overviews produce "because it alone has influence over the AI's offering and the algorithms with which the AI operates." Therefore, the Search giant is liable for the "independent, new, and substantive statements" generated for the AI Overviews.

Apparently, at the hearing, Google claimed that most users would know "that information generated with AI should not be blindly trusted," highlighting that AI Overviews include linked sources folks can check for themselves. The court rejected this argument on the grounds that the capacity to check claims made via AI Overviews does not "regularly exempt from liability for this statement."

To put it another way, if I were to write something heinously false about Google right now, the fact that you could probably very easily look elsewhere online to disprove my claim would not save me from the end of my journalistic career.

I'd rather not get into the specifics of how libel law works in the UK, so instead let me explain why this German case is also interesting when it comes to free speech protections for AI-generated statements. Specifically, the court wrote, "[An AI-generated statement is] not the expression of an acquired conviction of the persons expressing it, but the result of an algorithm."

I would not be surprised if similar reasoning starts to crop up in future legal cases internationally. The court also described AI-assisted research as "above all an expression of Google's business activities" and "at most a secondary expression of an interest in being able to freely express one's opinion and beliefs."

Long story short, the court has ruled that, though you can often easily fact-check what you read in an AI Overview, Google is still liable if this particular Search product makes false claims. As such, Google has been served with an injunction against disseminating false claims about the Munich-based publishers, and the company also had to cover 80% of the legal costs.

While this case is now concluded, I wouldn't be surprised if we see its ruling ripple across the international legal landscape. There's never a guarantee different legal systems will agree on the same arguments, though—and I can't help but wonder how this case might've played out Stateside.

Despite a rough couple of years for the company, Intel could be set to make a turnaround with both its handheld chips performing well, and its Foundry business taking on substantial orders, at least if recent reports prove to be true. It seems like the AI boom has really squeezed semiconductor leader TSMC's supplies dry, and Intel is now being tapped to provide some chips in its place.

As reported by The Information, Google has reportedly ordered at least three million chip orders from Intel, in the form of TPUs (Tensor Processing Units), according to two people familiar with the company. This Google-designed unit is specifically created for use in neural network machines. In case that sounds like gibberish, it's basically Google, yet again, further committing to AI.

These TPUs not only go towards developing Google's own AI services but will also be sold to companies like Apple and Meta. Google is reportedly expected to make more than six million TPUs between 2027 and 2028, so it will presumably be relying on both TSMC and Intel to provide the facilities to help make these chips.

TSMC is the world's leading semiconductor manufacturer, and Intel's biggest competition. So, in a sense, the ongoing memory crisis sucking up all the resources is somewhat beneficial to Intel, which may not have been able to make such a deal if TSMC were able to meet demand. That's before mentioning that the TPU demand wouldn't be so high without the AI boom.

As well as this, Nvidia is reportedly testing Intel to see if its chips can be used in its next major project, seemingly its next GPU architecture. Currently codenamed Feynman, it supposedly combines four graphics chips into one unit, so if Intel can win the contract, that could be just one more step in establishing the Foundry's foothold in the gaming market.

Nvidia is reportedly TSMC's biggest buyer, beating out Apple, which used to be at the top of the list, with the company providing the chips in all of its phones, laptops, and tablets. However, TSMC has been struggling to meet demand for a while now. It is now reportedly sold out until at least 2028, with even its Arizona fab being fully booked before it is even built.

Just last year, TSMC announced a $100 billion investment plan into three US-based fabs, but it's not the only one with expansion plans. Just last week, Intel seemingly started targeting a $3.3 billion factory in India. It seems both companies have big plans to ride the AI wave as far as it will take them.

One can only hope that this extra investment will contribute to stabilising supply sooner, rather than later. I can't hold out on upgrading my PC forever.

These days, it's hard not to view Google as a villain. The monopoly is currently engaged in a mission to dismantle the internet as we know it, scraping information from websites and turning it into frequently inaccurate AI nonsense. It's a mission that will likely kill off most of the websites it's feeding on. But it's not all bad. Google's also raising an army of 32 million mosquitoes that it's hoping to unleash upon America.

I know this sounds more like something an over-the-top videogame antagonist might be planning—like Metal Gear Solid's hornet-loving weirdo, The Pain—but actually Google is doing some good here by attempting to reduce the threat posed by the world's most dangerous animal, as reported by The Guardian.

Mosquitos carry a variety of diseases, not least of which is malaria. Around 700 million people a year are victims of mosquito-transmitted diseases, and around a million people die every year from their nasty little bites. When I lived in Zimbabwe, taking malaria tablets every morning became as normal and mundane as putting on my socks.

This is where Google's Debug project comes in. The aforementioned army of mosquitoes are exclusively male—which don't bite or carry these deadly diseases—and Google wants to unleash them upon California and Florida in two waves of 16 million. When they're out in the wild, they will mate with female mosquitoes, but due to a bacteria the males have been infected with, wolbachia, the females' eggs won't hatch.

"Over time," Debug states, "there will be fewer and fewer bad mosquitoes." The project also emphasises that this is a natural solution. "This technique uses a naturally occurring bacteria and uses no chemicals, no toxins and doesn’t involve genetic modification. Similar approaches have been used to safely combat other pests for decades. We’re combining the Debug team's scientific and engineering expertise with the help of international partners to raise and release lots of good bugs and stop bad mosquitoes that can spread disease."

This is all good stuff, and built on decades of research as well as the well-trodden sterile insect technique, which has already been used on other troublesome bugs. Google even found a positive way to use AI, which helps Debug separate male and female mosquitoes and release the males in appropriate locations and in the right numbers.

It's almost like AI could be a force for good if it wasn't so busy plagiarising content and having wild delusions.

As The Guardian notes, Debug has already done this in Singapore, initially working with the country's Nation Environment Agency to release 6 million male mosquitoes a week—the number has since risen to 10 million. This has resulted in a "70% reduction in dengue incidents after 6 to 12 months of releases".

This is genuine life-saving work, which makes it slightly harder to hate Google. But I think I can still manage it.

Privacy-focused search engine DuckDuckGo has announced that its opt-in, generative AI-limiting "No AI" search page has received three times as much traffic since Google's latest round of updates and PR for AI mode search. In addition to user privacy, DuckDuckGo is positioning itself as a destination for user choice—one that's still keeping its options open with AI.

Three days ago, we reported on how installs of DuckDuckGo's app were up by almost a third following Google's latest push for AI in search. This latest announcement from DuckDuckGo appears to exclusively pertain to noai.duckduckgo.com, its AI opt-out search option.

Since Google revealed its plans for an AI search overhaul, visits to our "No AI" search page have tripled…and they’re still rising! Want to make it your default on Chrome or Firefox? Grab our No-AI extensions and banish AI-assisted answers, chat, and AI images. (1/3)

— @duckduckgo.com (@duckduckgo.com.bsky.social) 2026-05-30T20:57:54.517Z

"Since Google revealed its plans for an AI search overhaul, visits to our 'No AI' search page have tripled," the company wrote on Bluesky. "And they’re still rising! Want to make it your default on Chrome or Firefox? Grab our No-AI extensions and banish AI-assisted answers, chat, and AI images."

DuckDuckGo also provided links for its browser extensions to make its AI-free search the address bar default in both Google Chrome and Firefox.

This popularity has followed a long push by Google for generative AI features in its ubiquitous search engine. While it hasn't yet replaced the classic 10 links results page, as some have suggested, Google's AI mode search has been placed in a more central location in the main search homepage, giving it primacy over traditional search in the UI.

This was accompanied by new public statements by Google CEO Sundar Pichai about AI adoption and its importance to Google, and follows years of creeping AI features at the forefront of the user experience, with AI Overview being a prime example.

It's worth noting that the No AI search doesn't stem from a principled anti-AI stance on the part of DuckDuckGo⁠—it's more a pragmatically benevolent centering of user choice. DuckDuckGo offers a corresponding AI maximalist experience on the other end of the spectrum with duck.ai.

It's a sensible, perhaps mercenary business decision I'm inclined to raise an eyebrow at but not necessarily lambast. As the sages of the Simpsons writers room once put it, "Abortions for some, miniature American flags for others."

DuckDuckGo's AI-free search is possibly the most extensive option when it comes to limiting your exposure to the tech on the web: In addition to lacking an AI overview, it seems to filter out AI text and image results.

But it's not the only game in town. There are Google Chrome extensions such as Bye Bye, Google AI which hide Google's own AI features. I've been kicking it old school: I set Chrome to default to Google's classic-style "Web" search, which is otherwise buried in the UI. I might consider a switch to something more extensive like DuckDuckGo, though, particularly if Google continues to sacrifice search's basic utility for its AI experiments.

ByteDance, the company behind TikTok and video editing app CapCut, is developing its own CPUs in a bid to better support its AI infrastructure. Given the AI industry's hunger for silicon, it's hardly surprising another massive tech company is taking hardware into its own hands.

Several external partners have already been approached with regard to design work for ByteDance's chip. Securing capacity at manufacturing foundries was apparently also discussed. The plan is in the early stages and not yet public, but Reuters spoke with a number of anonymous sources familiar with the matter.

Currently, ByteDance is exploring two different chip architectures: Arm and RISC-V. Both architectures are used for chips in data centres throughout the wider industry, though Arm presents a proprietary ISA, with a fixed feature set to go along with it, whereas RISC-V is a modular, royalty-free, open-source architecture. I can imagine not having to pay Arm's licensing fees may be especially appealing to any major player looking to make their own chips.

There's been a lot of excitement about Arm chips in the realm of PC gaming, especially as the hardware team waits for even a whisper of news about Nvidia's still-yet-to-be-announced N1X Arm chip (various reports have insisted it's coming at some point this year). As for the alternate, open-source architecture…well, a few years back someone built a fully compliant RISC-V computer inside Terraria—and then they played Pong on it.

As the AI industry reaches towards CPU-intensive 'inference', ByteDance's planned chip is intended to be deployed throughout its AI data centres and servers. This, in turn, will support ByteDance's agentic AI products, such as the company's development platform, Coze.

Reuters reported earlier this year that Intel had warned its Chinese customers that orders for processors could take six months to fulfil. The same report claims AMD was looking at 10-week delivery times in the same market, and ByteDance currently sources its CPUs from these two partners.

With that in mind, alongside the knock-on effects of the memory supply crisis, it's unsurprising ByteDance would take the route of making its own hardware. It's not the only company to go this way, either, with Google announcing its own Arm-based CPU all the way back in 2024. Microsoft, too, is moving towards designing more custom silicon for its own data centres, and so is Amazon.

As much as I enjoy a gacha pull as a little treat, I'm not overly jazzed about how inescapable gambling mechanics have become in games—or anywhere else, for that matter. So-called prediction markets are also becoming increasingly popular, and one Google engineer has been arrested for allegedly leveraging insider information to place bets on Polymarket, a cryptocurrency-based platform.

Michele Spagnuolo was charged this week on the grounds that he'd used confidential business information to place a number of bets on Polymarket (via Wired). Up until his arrest, Spagnuolo had worked for Google since 2014, most recently as a security engineer at Google Zürich. His Google research page has since been taken down.

FBI agent Brandon Racz wrote in the legal complaint, "Unlike the counterparties to his trades, Spagnuolo knew the outcome of these wagers before the trading public did because he had accessed Google’s confidential, commercially valuable internal data."

For those unfamiliar, prediction markets encompass more than sports betting, allowing users to wager on the outcome of a variety of future events—such as whether GTA 6 will get delayed again. More Perfect Union's recent reporting on Polymarket makes for illuminating viewing (and, in my humble opinion, offers plenty of reasons to never go all-in on prediction markets).

Spagnuolo, going by the Polymarket handle 'AlphaRaccoon,' allegedly bet on who Google’s most-searched person of the year would be. He netted $1.2 million in total, after correctly predicting that the winner would be D4vd, the musician charged with first-degree murder in April this year.

How Spagnuolo allegedly sought and leveraged the insider information is somewhat vague at this point in the case. A press release about the case from the Southern District court of New York attempts to explain, "In connection with his role, Spagnuolo had access to Google’s internal data systems, including an internal software tool that provided him with access to confidential, nonpublic data. That software tool bore a banner that stated, in part, 'Google Confidential' in red text. Indeed, Spagnuolo certified his understanding of various Google confidentiality and ethics policies."

Spagnuolo has been charged on multiple counts, including commodities fraud, wire fraud, and money laundering. These charges present a potential maximum prison sentence of between 10 and 20 years.

It's important to stress that Michele Spagnuolo has only been charged at this point (and charges do not indicate someone is guilty inherently). Though Spagnuolo was arrested in New York, it's also worth noting he is an Italian citizen working for Google Zürich, which could complicate any future trial that may take place.

"Today’s charges reinforce a decades-old message: corporate insiders cannot use confidential business information to turn a profit in our markets," US Attorney Jay Clayton said, "As alleged, Spagnuolo violated the duties he owed to his employer and used Google’s confidential business information to make more than $1.2 million in trading profits on Polymarket. Insider trading compromises the integrity of our markets, and the American people want this greed-driven conduct investigated and prosecuted."

As we march ever further into our bright AI future, it's somehow reassuring to know the tech can still stumble at the first hurdle. Or concerning, one of the two. Google Search users will have noticed that the AI Overview feature has been upgraded recently as part of the company's efforts to crowbar more AI into, well, everything.

What this means in practice is more conversational, LLM-generated responses to basic queries—but they still sometimes reveal more about the tech's capabilities than Google would like.

Techcrunch has noticed an old LLM favourite appearing within AI Overview's ever-more-prominent top box: the inability to correctly identify letters within words. A simple search query of "How many Ps are in Google" can cause an... inaccurate response.

"There are 2 'p's in the word Google," the AI merrily responds, before guessing the query might be related to an expression of the mathematical number googol. I tested this myself and got the same answer on Chrome, although it disappeared later. It still "works" on Firefox at the time of writing, though.

Anyway, it's far from the only spelling question AI Overview currently struggles with. When I asked how many Rs are in the word "enigmatic" the AI confidently answered that there's, err, one. Bizarrely, the response then correctly spelled out the word using individual letters. With, of course, no Rs.

This reveals a basic truth about how LLMs work. Words and letters are represented by tokens within transformer-based models, which means the AI doesn't "read" the word the same way you and I do. The text has been converted into numerical representations, which are then contextualised.

Speaking to Techcrunch, AI researcher and assistant professor at the University of Alberta Matthew Guzdial said:

"LLMs are based on this transformer architecture, which notably is not actually reading text. What happens when you input a prompt is that it’s translated into an encoding. When it sees the word 'the,' it has this one encoding of what 'the' means, but it does not know about 'T' 'H' 'E.'"

Google told the outlet that "counting within words had been a known challenge for LLMs, and we're working to fix this particular issue." Which, based on my earlier testing, may mean disabling AI Overview responses to certain queries while it figures out a solution.

All of this has come up before, of course. The inability of AI tools to correctly answer certain requests or spell certain words is a well-known phenomenon. However, putting more LLM-based responses right at the top of user's daily search queries seems to have created more issues than Google may have hoped for.

And then there's the potentially massive compute demand. Writing simple phrases into the search bar now often results in an AI response that seems, at best, like an inefficient use of resources. I typed "I admire your bravery" into the Google search bar this morning, and received the following response:

"I'm flattered, but I'm just an AI! It's the humans tackling big challenges, creating things, and pushing boundaries who are truly brave. I'm just here to make things a little easier for you.

"What can I help you figure out or conquer today?"

Have you ever watched a YouTube video and immediately screwed up your face in an attempt to discern whether it's real or AI-generated? Welcome to a daily part of my life. The good news is that YouTube has listened to user feedback, and as a result is moving AI disclosure labels for photorealistic and "meaningfully AI altered or generated" content to a "more prominent" position.

The bad news is, the labels look pretty tiny in the promo video. The AI warnings will now sit directly below the video player for long-form AI-generated videos, above the description. For short form, it's an actual overlay on the video itself. Which is indeed, a prominent position.

But yes, both versions still look like they'd be easy to miss. I suppose a big honking "THIS VIDEO IS BULL****" with red flashing warning signals on either side would be a bit much, but I'm not sure it does much to assuage fears that AI-generated videos are becoming harder to spot.

What might help a little more in this regard, though, is YouTube's other announcement—that "new internal signals" will be used to help identify AI-generated content.

YouTube creators are required to manually disclose when they use realistic AI, but I'm guessing that a fair few of them... don't. With the introduction of this tech, though, the platform says it will be able to automatically apply a label to anything with "significant photorealistic AI use."

"As this technology continues to improve, creators remain in control," says YouTube. "If a creator thinks their content was incorrectly identified as AI-generated, they can update the disclosure status in YouTube Studio.

"However, disclosures will remain permanent in a handful of cases, including: Content created using YouTube's own AI tools, like Veo or Dream Screen [and] content containing C2PA metadata indicating they were fully generative AI."

"Our goal is simple," the blog post continues. "Make it as easy as possible for creators and viewers to have the right information."

A noble goal, to be sure. However, in a world where YouTube dominates the online video market (and amid real concerns about the level of AI slop on the platform), this new labelling system feels like small potatoes. Very small potatoes indeed, come to think of it.

Efficiency gains are often cited as the reason to bring more AI into the workplace, though concerns over resulting job losses are never far behind. It might surprise you, then, to learn that the head of Google DeepMind, Demis Hassabis, actually thinks job cuts caused by AI are not the way forward.

Today's wealth of machine learning tools means that anyone can vibe code—including this dog. Still, Hassabis thinks it's premature to call time on software development as a profession. “I have no idea why people are going around talking with certainty about that,” Hassabis told Wired this week.

In that same conversation, he later went on to criticise companies looking to replace engineers with AI, saying, "I think it's a lack of imagination—and a lack of understanding of what's really going to happen."

So, what future does Hassabis envision instead? Rather than efficiency gains at the cost of livelihoods, Hassabis instead asks why not keep the efficiency and the engineers; he says, "From my point of view, from DeepMind and Google's point of view, if engineers are becoming three or four times more productive, then we just [want to] do three or four times more stuff."

To me, that sounds a lot like 'doing more stuff with the same amount of resources' and not necessarily something ultimately sustainable, but then I'm not a software engineer. "I have a million ideas, from lab drug discovery to game design,” Hassabis continues, “I'd love to have some free engineers to go and do those kinds of things.”

You'll recall that many moons ago, Hassabis worked in the games industry, spending a number of years at Lionhead Studios before moving on to found Elixir Studios in 1998. Hassabis designed and directed 2003's Republic: The Revolution, and also worked on 2004's Evil Genius, though the studio closed in 2005.

Hassabis also reflected on the subject of games and how today's AI has yet to create a killer title without human help. He said, "I think there's something missing."

Last year, Google DeepMind's own Genie 3 could generate an interactive world that ran at 720p, 24fps, and only remembered what you did for one minute. The model has come a ways since then, with the public preview version now simulating physics and generating the path in front of your player avatar in real time.

That said, The Verge noted in their hands-on that Genie 3's overall output remains "much worse than an actual handcrafted video game or interactive experience." Not to bang on about the nebulous concept of 'human creativity', but I think it's clear that 'something missing' may never be in the grasp of a generative AI model's ability.

As a writer of news on the web, there's a lot I could say about the integration of AI in Google Search. Instead I'll just say this: I've had Bye Bye, Google AI—a Chrome extension that hides Search's AI Overviews among other things—installed since our Jeremy wrote about it last year. But according to Google itself, my personal disdain for its AI offerings is something of an outlier.

Just for a start, Google CEO Sundar Pichai says, "AI Mode has been a revelation, our biggest upgrade to Search ever. People love it, and in just a year, it’s already surpassed 1 billion monthly active users."

Those AI Overviews that once annoyed me so much apparently enjoy over 2.5 billion active users every month. Pichai goes on to elaborate, "When people use our AI-powered features in Search, they use Search more. Search has become less about individual queries and feels more like an ongoing conversation, giving you deeper insights and connecting you with the vastness of the web."

According to Pichai, it's not just Search's wider AI functionality seeing success, though. The Gemini App is incredibly popular too, apparently going from "400 million monthly active users" last year, to over 900 million now. It's worth noting that Gemini saw the introduction of Personal Intelligence around the start of the year. This is an opt-in feature that allows Gemini to sift through your personal data across Google's apps for "more customized and helpful" responses. Grand.

Gemini leads the charge for 'natural, conversational AI' in a range of Google's products. The model can now field queries via Ask YouTube, and Ask Maps, as well as facilitate 'brain dumps' in Google Docs. Pichai elaborates, "To create a doc with Gemini before, you had to type out a precise prompt. With Docs Live, you can just verbally 'brain dump' whatever is on your mind, and let Gemini do the rest."

Across all of Google's consumer and developer-facing AI products, the company's models are said to be processing over 3.2 quadrillion tokens every month—and that requires serious infrastructure. To that end, the company expects to spend between "approximately $180 to $190 billion" in capital expenditure. Custom silicon is a big part of that investment, with Pichai specifically highlighting Google's AI chips TPU 8t and 8i as "more energy efficient, delivering up to two times better performance-per-watt."

AI is power-intensive, with Google's own emissions up by 51% last year. But those latest energy efficiency claims are perhaps little comfort in light of one recent report suggesting gas power projects for just 11 US data centres could emit more greenhouse gases than entire countries.

But allow me to set my climate anxiety to the side for the moment—I have mixed feelings about the purported popularity of Google's AI products. On the one hand, it's not hard to see the appeal of a conversational interface for a perhaps less tech-literate user base. On the other hand, I'm intrigued by the choice to measure the success of these AI products by metrics such as user count, tokens processed, and capital expenditure rather than, say, profit.

I'm also curious about how Google is defining 'active users'; how many of those millions genuinely feel like they're getting the most out of Google Search's AI Overviews, and how many are simply unclear on how to avoid it? Presently, beyond third-party extensions like Bye Bye, Google AI, there's no official way to opt-out of those AI Overviews.

Maybe the dollar signs I'm looking for are on the way. Still, one uncharitable way to read much of the above would be as a reassurance to, say, investors that there is enough demand for these free products that a significant enough portion of the user base will convert into sales for paid AI products down the line.

If users aren't willing to eventually pay for AI products more broadly speaking, the bubble may finally burst. At the risk of paraphrasing recent comments made by US senator Elizabeth Warren, it seems to me that revenues from AI products aren't necessarily in step with the huge capital expenditure required to fuel them. "I know a bubble when I see one", she said. Quite.

We reported on the booing of a speaker at the University of Central Florida last week, after they declared AI to be "the next industrial revolution" in front of the assembled students. However, that's nothing compared to the absolute barracking ex-Google CEO Eric Schmidt received when speaking at the University of Arizona's 2026 commencement ceremony last Friday.

Addressing thousands of students at the Casino Del Sol Stadium, Schmidt began his speech by discussing the impact of modern technology on the world. "In a sense, we thought that we were adding stones to a cathedral of knowledge, that humanity had been constructing for centuries," Schmidt opined, as the crowd began to rumble underneath him.

It was the mention of AI, however, that really got the audience going. "Last December, Time magazine selected its person of the year for 2025, and this time it was the architects of artificial intelligence," said Schmidt, which caused a large number of the crowd to erupt into jeering and boos.

"So today, we stand on this edge of another technological transformation," Schmidt continued. "One that will be larger, faster, and more consequential than what came before. It will touch every profession, every classroom, every hospital, every laboratory, person, and every relationship you have."

Unsurprisingly, this doubling down on the impact of AI resonated poorly with the crowd. Responding to the jeers, Schmidt said: "I know what many of you are feeling about that. I can hear you. There is a fear..."

Schmidt paused, as the boos once again reached fever pitch. "There is a fear in your generation that the future has already been written. That the machines are coming. That the jobs are evaporating. That the climate is breaking. That politics is fractured. And that you are inheriting a mess that you did not create.

"And I understand that fear," Schmidt smiled, holding up his hands in an apparent act of appeasement. "It's rational. And it's amplified every day by social media platforms with algorithms that have learned with great precision that fear earns clicks and anxiety drives engagement."

Schmidt then tried to strike a hopeful tone: "I want to say something to you this evening, as clearly as I can. To speak of the future as though it has already been decided is to surrender the one thing that actually matters. You're surrendering your agency."

"The question is not whether AI will shape the world... it will. The question is whether you will help shape artificial intelligence."

Louder boos, more jeers, more nervous smiling from Schmidt. If I may be so bold, it may not have been the best idea to list the legitimate concerns of a stadium full of young people regarding their grim-looking AI futures—before handwaving them away as social media brainwashing that they can resist, actually, if they get onboard the AI train.

After listening the potential benefits of AI in the fields of science and medicine, the biggest boos of the evening came when Schmidt declared:

"If you don't care about science, that's okay... because AI is going to touch everything else as well. Whatever path you choose, AI will become part of how work is done."

It's a bit like watching every public speaker's pre-show anxiety dream come to life. However, the full cause of the boos may not have been entirely down to the student's rejection of their new AI-influenced world. According to Business Insider, some attendees had planned to boo Schmidt over previous sexual assault allegations.

And while Schmidt began being booed before he even reached the lectern, his comments on the AI future his ex company—and companies like it—are foisting upon us all certainly didn't make anything any better.

While there's no chance that it was developed because Apple had chosen to do something similar, one can't help but draw parallels to the MacBook Neo. The newly announced Googlebook has your phone and AI at the heart of it all, and will supposedly bring "the best of Android…and ChromeOS", utilising hardware from Intel under the hood.

That's pretty much it as far as details go, other than some swooping imagery of glossy-looking laptops, a short tweet from Intel on the collaboration, and a statement about "working with industry leading partners like Acer, ASUS, Dell, HP and Lenovo."

Naturally, Gemini AI is going to be at the forefront of what the Googlebook can do, but if you have an Android phone, you'll be able to cast apps and access files directly from the phone without having to wire things up or install them. How well all this will work in reality is anyone's guess at this stage, but if it works as seamlessly as suggested, a Googlebook might be a good choice of laptop for work-on-the-go.

The real question of importance, however, is precisely what hardware is going to be underneath the hood. Since Intel is involved, the most obvious pick is either a small Panther Lake chip or even smaller Wildcat Lake. Both options are going to be very nice to have, though they really need fast LPDDR5 memory to shine.

Given how expensive DRAM is at the moment, if Google takes a leaf from Apple's MacBook Neo and goes with something like 8 or 12 GB, then it's unlikely that you're going to see any Googlebook topping a benchmark chart. That said, Wildcat Lake is light on power demand but strong on single-core performance, so you never know.

Perhaps the biggest challenge facing the Googlebook is the legacy of Chromebooks. These have been around for many years now, and while generally very cheap, they've accrued a reputation for being slow and of limited capability. Google's new devices are unlikely to be at the bottom end of the price scale, but the spectre of the Chromebook is going to be hard to exorcise.

We'll know for sure when the first models appear later this autumn.

I don't doubt there's some worthwhile application of at least one technology lumped under the monolithic 'AI' banner. Unfortunately, big tech's major players seem preoccupied with reinventing the wheel and introducing an unnecessary agentic twist. Case in point, Google DeepMind's latest experimental demo attempts to give your humble mouse cursor an AI enhancement.

"The mouse cursor is something that has been forgotten," argues Adrien Baranes, a staff researcher prototyping Human-AI Interactions at Google DeepMind, "What if behind the pointer, there was an AI model, like Gemini, trying to interpret whatever we are saying, like another person would."

To be fair to the demo, barking commands at Gemini does reduce the standard, laborious process of copy and pasting recipe ingredients into a shopping list by a considerable number of clicks. I've got plenty more sass where that came from—but this cursor demo is also interesting in how it attempts to navigate contextual challenges that might otherwise stump many AI models.

Essentially, rather than relying on an AI model to consistently tell the difference between a shopping list, a recipe, a food fight, and a hamburger costume, a combination of cursor gestures and naturalistic commands like 'move this here' are leveraged to point the AI in the right direction.

"Current models require precise instructions, but our AI-enabled pointer removes that burden," Google DeepMind shares, "By 'seeing' what’s under your cursor, it instantly understands the specific word, image, or code block you need help with."

Another example explored by these early tech demos involves watching a video of 'top 10 places to eat in Tokyo,' dragging your cursor across an eatery's signage, and then Gemini agentic-ly taking you through booking a table for the following evening. Setting aside the well-covered security concerns of letting an AI agent at your emails or other important data, I do also wonder how this tech might handle misclicks—at least with the restaurant example, there appear to be multiple steps that a user can easily backpedal away from.

Otherwise, I'm not convinced this "50-year-old interface" really needed the AI reinvigoration. Besides the obvious 'if it ain't broke…' argument, I don't think I'd be comfortable with allowing Gemini to get an eyeful of my desktop.

To be clear, if you switch Smart Features on in Gmail and let Gemini organise your inbox, Google won't then scan your emails to train its AI. Instead, official support documentation for Gemini apps says that "summaries, excerpts, generated media, and inferences" resulting from your prompts to the AI are what is used as training data.

As such, if the cursor demo was to become more widely available, odds are Gemini wouldn't be tattling to Google about the contents of your SSD—though it could potentially tell Google what you do all day at your desk. Personally, I'd rather no one knew how often I fail to write 'embarassment' or 'occassionally' correctly, let alone anything else.

I've been keeping an eye on the various documents released as part of the ongoing Musk vs Altman legal fight, which centres around OpenAI changing to a for-profit structure and Elon Musk's belief he was deceived by Sam Altman. The documents, which consist mainly of emails and text exchanges between the various figures involved, turn out to have several themes running through them: and an absolutely huge one is the co-founder and CEO of Google DeepMind, Demis Hassabis.

Hassabis is widely regarded as the outstanding talent in the AI field, a fellow of the Royal Society, and in 2024 shared the Nobel Prize in Chemistry with John M. Jumper for their work on AI protein structure prediction. He also began his career in videogames at Bullfrog, before working as lead AI programmer on Lionhead's Black & White, and founding his own developer, Elixir Studios (which made Republic: The Revolution and Evil Genius).

But we're talking about Hassabis today because Elon Musk and his various AI buddies have a real obsession with the dude, which at times seems to veer into the unethical. The question really is whether it's Hassabis' trustworthiness that bothers them, or the fact that they don't want Google to create and thus have theoretical control over artificial general intelligence.

Back in February 2016, Musk emailed Sam Altman and Greg Brockman about hiring at OpenAI. "We need to do what it takes to get the top talent," writes Musk. "Either we get the best people in the world or we will get whipped by Deepmind. Whatever it takes to bring on ace talent is fine by me."

Musk goes on to outline what he believes to be the core issue: "Deepmind is causing me extreme mental stress. If they win, it will be really bad news with their one mind to rule the world philosophy. They are obviously making major progress and well they should, given the talent level over there."

Musk and others put that "one mind to rule the world philosophy" squarely on Hassabis' shoulders, and a few years later comes an even more eye-popping exchange.

"I think there are a lot of no-brainers to explore, and I put some in that doc, but the thing that keeps calling out to me is there is a very low probability of a good future if someone doesn't slow Demis down," writes Shivon Zilis to Musk [PDF] on 16 February 2018.

"Slowing him down is the only non-negotiable net good action I can see. You don't realize how much you have an ability to influence him directly or otherwise slow him down. I think you know I'm not a malicious person but in this case it feels fundamentally irresponsible to not find a way to slow or alter his path."

I mean, that's like a mob boss talking, capiche? Musk responds: "Best to talk by phone about this later tonight. I doubt I could do so in a meaningful way."

"OK, yes that would be good," says Zilis. "And, ultimately up to you of course, but I really think you can so would like to at least make the case. In any case, I will sleep better at night for having tried!

"Will leave you be on the Demis stuff. I'm sure it's hard to think about and you have so much on your shoulders all the time that I always feel terrible pushing you. I just needed to say it once since it's been plaguing me."

One intriguing message from Zilis to Musk [PDF] has, unfortunately, been redacted, though again shows how he's always uppermost in their thoughts: "Will leave you be on the Demis stuff. I’m sure it's hard to think about [redacted]. I just needed to say it once since it's been plaguing me. [redacted]."

Later in 2018, Musk is back on Hassabis, and launches a broadside at Altman and Brockman about OpenAI's relative position.

"My probability assessment of OpenAI being relevant to DeepMind/Google without a dramatic change in execution and resources is 0%," writes Musk in late December. "Not 1%. I wish it were otherwise.

"Unfortunately, humanity's future is in the hands of Demis."

Musk then links this NYT article about the success of Deepmind's AlphaZero chess software, before continuing:

"And they are doing a lot more than this.

"OpenAI reminds me of Bezos and Blue Origin. They are hopelessly behind SpaceX and getting worse, but the ego of Bezos has him insanely thinking that they are not!

"I really hope I am wrong."

A few days later Musk is back on his Hassabis hobbyhorse.

"OpenAI is not a serious counterweight to DeepMind/Google and will only get further behind. It is surprising that this isn't obvious to you," writes Musk. "In general, always overestimate competitors. You are doing the opposite."

But Hassabis also weighed heavily on the minds of others. Satya Nadella gave pre-trial testimony in September 24, 2025 [PDF], and one of the lawyer's questions is "Google was—I believe you said by far the dominant player in machine learning around 2015?"

"I would say so, yeah," replies Nadella, before the lawyer mentions Google's acquisition of DeepMind and asks whether Microsoft believed the company was "making a lot of progress in that field of machine learning?"

"I now can't recall specifically what state DeepMind's breakthroughs were, but yeah DeepMind was well known, even in that time frame, and Google had DeepMind, had Google Brain. They had many, many different efforts they were publishing."

Nadella's testimony can be condensed into his acknowledgement that Google led the AI race from the 2010s onwards, and that Microsoft was tracking DeepMind's progress. Nadella says "I probably knew [Hassabis] a little" from around 2015 "just after I became CEO."

The lawyer drills down and asks why Nadella and Microsoft were tracking DeepMind specifically.

"Just because of the breakthroughs that this particular regime of AI around deep neural networks were showing real promise of making breakthroughs in fields like language translation that had not been seen before. And so that's why we were waiting to see how we could also participate and make sure that we have those breakthroughs."

Amusingly enough, the lawyer moves on to OpenAI and Dota 2. One of OpenAI's first goals was to create a bot that could beat humans at Dota 2, a task in which it eventually succeeded: And as PCG reported last week, that was because Elon Musk had personally called Microsoft CEO Satya Nadella to secure a massive discount on access to Azure, the company's cloud computing platform.

Here, the lawyer begins by asking what Nadella understands about the game.

"I'm not a gamer," says Nadella. "And I think it's a Steam game, if I'm not mistaken." He then goes on to give a brief overview of why he thought the game angle with AI was interesting.

"That's why gaming environments being closed worlds were a great sort of, you know, environment to do reinforcement learning," says Nadella. "Right? The objective function is clear. The reward function is clear.

"I forget now when and what time frame some of the breakthroughs on AlphaGo and so on happened but, you know, Demis was a game developer. There's a long history of AI developers who came out of using games in environments, building AI bots in games, so it's sort of a given."

Elon Musk also gave pre-trial testimony a few days later, on September 26, 2025 [PDF], in which he reveals that poaching people from DeepMind caused him to have a fallout with Google co-founder Larry Page.

"I talked to dozens and people over the years," says Musk. "But I think the most crucial recruit was Ilya Sutskever. In fact, the recruitment of Ilya was what actually caused Larry Page to stop being friends with me.

"So Ilya went back and forth multiple times saying he would join OpenAI or stay at Google, and ultimately agreed to join; and Larry Page and Sergey [Brin] and Demis Hassabis did everything they could to keep Ilya. When Ilya finally decided to join OpenAI, that's what ended the friendship with Larry Page. He didn't talk to me after that."

The lawyer asks if Musk is still getting the silent treatment a decade later, and it's a simple "yes" before adding "they were very upset about [Sutskever]."

Let's end on the funny note of Sam Altman's delusions of grandeur. This document [PDF] was filed on 6 January 2026, but unfortunately is not otherwise dated. Based on what he's saying though I would date it in the region of 2016-2018.

"Progress fundamentally has to be made by non-profit, interesting direction you could go," writes Altman. "Everything I perceive with OpenAI, race dynamics vs Demis + brain + whatever, gotta get there first."

I'm not quite sure but "brain" may be a reference to neural chips, though of course what's amusing here is that once again Hassabis and the "race dynamics" of competing with him is living rent-free in Altman's head. Then, the ego really takes over.

"Another angle: [DeepMind will] never do that much that's interesting," writes Altman. "It is better for us to become increasingly kings of this industry. The choice defines us.

"You say we should have been more ok with giving it to Elon. While it's true, it's also the case he's now given it to us. The grand upside is I want it. Need to stop letting distractors get to me/us. Being the Kings of AI is not so bad."

The Kings of AI! Good band name. Also: Jesus wept. Do we really want this dweeb potentially making big decisions about humanity's future?

In early 2019 Altman is still trying to tempt Musk into phone calls by promising "some mild Demis updates to share" [PDF] while in 2023 Mira Murati is emailing Nadella [PDF] saying "it is very important that we don't lose researchers to Demis or Elon."

Let's end on one of the least-crazy things anyone says about Hassabis, and it's from Ilya Sutskever [PDF] who actually worked with the guy.

"The goal of OpenAI is to make the future good and to avoid an AGI dictatorship," Sutskever writes to Musk in September 2017. "You are concerned that Demis could create an AGI dictatorship. So do we.

"So it is a bad idea to create a structure [for OpenAI] where you could become a dictator if you chose to, especially given that we can create some other structure that avoids this possibility."

It's a great point: but answer came there none.

The Associated Press has a new report on a cafe in Stockholm that's being used for an AI experiment, in which an AI bot using Google Gemini runs the show while all the actual, y'know, coffee-making is done by humans at its command.

The firm behind it, Andon Labs, was previously involved in an experiment where an AI ran a vending machine, and proceeded to start selling stuff at a loss, before inventing fake people and meetings, then collapsing into a bizarre identity crisis. So this should be good.

Andon Labs' bot is an "AI agent" called "Mona" but I am going to just call it the bot. The firm has self-effacingly called the new place Andon Café, and the idea is that the bot oversees all of the management-y side of the business: initially this was stuff like securing the proper permits and hiring staff, but the day-to-day task is ordering appropriate inventory and managing those staff.

Who, and I might repeat this again later if I feel like it, are the ones right in front of the customers brewing the actual coffee that they order. Right there. Like, a few feet away from you. A human being working in the service industry who wants to take my coffee order in a courteous fashion! Won't someone save me from this hideousness.

Alright. There are obviously good reasons to be doing thi… you know what, I can't finish that sentence. Speak ur brains, AI evangelists:

"AI will be a big part of society in the future, and therefore we want to make this experiment [to] see what ethical questions arise when we have AI that employs other people and runs a business," said Hanna Petersson of Andon Labs.

Petersson says the bot was given some basic prompts: run the place profitably, be nice, and figure out the nuts-and-bolts itself but ask for help when required. It proved efficient at arranging utilities and securing permits, and posting ads on job sites, but when it got down to the nitty-gritty of managing a customer-facing business things started to go a bit pear-shaped.

First of all, the AI doesn't respect working hours: it uses Slack (an instant messaging platform) to communicate with the baristas, but would "often" message them when they weren't working. This is frowned-upon in most sane countries, but in Sweden? Nej.

The bot arranged commercial contracts with bakeries, but then proceeded to screw these up on the reg: sometimes ordering way too much bread, and then sometimes not putting in an order at all. So… the cafe just had no sandwiches on those days.

It's going great! The bot also managed to order 6,000 napkins, 3,000 rubber gloves, and four first-aid kits for what the AP calls a "tiny cafe" alongside a load of canned tomatoes that aren't used in anything the place actually sells.

Is the AI making money? Nej. A thousand times nej. It has made $5,700 in sales since opening in mid-April, but started with a budget of "$21,000-plus" per Andon Labs.

Chalk another one up to the visionaries of the future. In a previous life I managed a bar, and somehow did the orders, checked the inventory, managed the rotas, and even served some drinks while I was at it. I'm not even saying I was good at it (I was good at it) but my biggest mistake was ordering twenty bottles of Grenadine, and at least you can eventually use that, unlike thousands of rubber gloves.

"When old memory of ordering stuff is out of the context window, [the bot] completely forgets what she has ordered in the past," says Andon Labs' Petersson. Yeah, you can see why that might be an issue. Maybe at some point we'll all wake up and smell the coffee—the coffee that has been made for us by human beings who can handle being told "I'll have a regular latte please."

Google's Chrome browser has been quietly downloading a 4 GB local AI model onto user's devices without asking permission. What's more, if you manually delete the model, Chrome will simply download it again.

It was security researcher Alexander Hanff, who runs the ThatPrivacyGuy website, who discovered Chrome's dubious behaviour. He found a weights.bin file measuring around 4 GB stored in Chrome's local AppData folders. As the filename implies, it's a weights file for Google's Gemini Nano AI model. And as Hanff notes, it is downloaded without the user's permission.

"It is the weights for Gemini Nano, Google's on-device LLM. Chrome did not ask. Chrome does not surface it. If the user deletes it, Chrome re-downloads it," Hanff says.

As he explains, "Chrome uses it to power features Google has marketed under names like 'Help me write', on-device scam detection, and other AI-assisted browser functions."

Regarding the permissions issue, there is no explicit checkbox in Chrome Settings for the model download. It's part of Chrome's broader AI functionality, which is enabled by default where present. This is where things get a little tricky.

By Google's own admission, AI features are enabled according to the capabilities of a given device. I understand at least 16GB of memory is required, for instance. However, a poll of the PCG massif revealed very mixed results. Most of us do not have the weights file, even on devices that likely meet the hardware requirements, while some of us have no sign of AI functions and features at all.

Google's response to Hanff's report reveals that this has been going on for some time. Indeed, back in February, Google says it added a option in Chrome settings to disable AI features, which in turn will prevent the model from being downloaded.

Long story short, it seems that Google has been rolling out this functionality to a limited subset of Chrome users with machines that meet the hardware requirements. How many isn't known and it's just one of several elements that lack transparency.

Indeed, apart from the obvious lack of user permission and those transparency issues, there are plenty of other problems. 4 GB is a significant chunk of data, both in terms of local storage and also bandwidth. Anyone with a metered internet connection, for instance, really needs to know about 4 GB being downloaded.

There's also the energy footprint of this kind of roll-out. Hanff has calculated that if this AI model were pushed to one billion users, the distribution of the data would guzzle 240 gigawatt-hours of energy and generate 60,000 tons of CO2 equivalent. Note, that's just getting the model out onto devices, never mind any energy used by the models locally on said devices.

As per the screenshot we've uploaded from our own Nick Evanson's machine, if your PC does have the weights files, it'll be located in subfolder of the local AppData files for your Chrome installation.

Similarly, you should find a toggle switch for "On-Device AI" in the System subsection of Chrome's setting menu. As I understand it, if you don't see that toggle switch it's because your machine either doesn't meet the hardware requirements or, if it does, you haven't been included in the roll out.

In response to Hanff's research, Google Chrome VP and GM Parisa Tabriz also posted on X explaining that, "on-device AI is core to our developer and security strategy."

She further revealed that this has been ongoing since 2024 and said, "while this requires some local space on the desktop to run, the model will automatically uninstall if the device is low on resources." Notably, Tabriz said nothing about the issue of user permission.

Overall, this is a pretty unsatisfactory, if not an entirely surprising, affair. Few if any users would assume that downloading a 4 GB AI model is a standard part of a mainstream web browsing package. So, getting permission seems like a no brainer. That Google thought this is all fine—and apparently still does—doesn't reflect terribly well on either the company itself or the attitude of the AI industry to safety and privacy.

For the record, we understand Chrome downloads the model to all three of the major desktop-class operating systems—Windows, MacOS and Linux—but not to mobile devices running Android and iOS.

For the banishment of all doubt, no, Google isn't scraping the contents of your email inbox in order to train its AI models. However, that doesn't mean Gmail is free of Gemini, and sooner or later you'll need to decide your own personal boundaries with AI.

Google announced at the start of the year that Gmail would be entering the Gemini era. This ushered in AI overviews for your inbox, as well as the Help Me Write feature that allows floundering users to draft emails with AI assistance. Gemini is used to sift through all the data in your inbox in order to summarise information or give relevant insights.

These features have been available for some time now, but the presence of AI in your inbox may have caught your attention again recently due to viral posts such as this one from Shark Tank's Lori Greiner (via Forbes). What you may have heard less about is 'Personal Intelligence'.

Personal Intelligence began to roll out around the same time as Gmail's Smart features, and allows users to link Gemini to various Google apps, like Gmail, Google Photos, and YouTube. You could ask Gemini 'Hey, why can't I stop thinking about low-poly rats?' and, in theory, Gemini could look through your YouTube watch history and messages, then say something like, 'It's because you rewatched Rat Movie: Mystery of the Mayan Treasure last week after your best friend sent it to you with the following caption: Rat squad 4 lyfe.'

Much like Gmail's smart features, when personal intelligence is enabled Gemini can search through all of the personal data held by Google's apps in order to answer your conversational queries. It can also scan all of your Google photos in order to "create more relevant, personal images using Nano Banana" (via Forbes).

It's worth reiterating that 'personal Intelligence' is an opt-in featureset. You can also opt-out of Gmail's smart features fairly easily if you'd rather Gemini didn't go rifling through your inbox. First, open Gmail and click the big cog icon in the top-right corner. On the Quick settings menu, click 'See all settings.' Now, scroll down to Smart features and untick the associated box. This does mean you lose access to a number of AI-assisted features, including automatic email categories in your inbox.

In the case of both 'Personal Intelligence, and Gmail's AI features, Google assured last month that its AI isn't snaffling up any of your data. This is in line with Google's assertion last year that it's not using Gemini's Gmail integration to train its AI.

That said, no inbox is really an island. Google says in support documentation for Gemini apps, "When you interact with Gemini, summaries, excerpts, generated media, and inferences from your relevant media, emails, and files may be used to help us answer your prompts. To make our responses relevant, helpful, and high quality, we train our generative AI models off of these summaries, excerpts, generated media, and inferences."

In other words, Gemini won't steal your photos or emails for training, but it will retain the conversations you've had about them and elements of that data may then be used to train the model. Even with this degree of separation, I still wouldn't want to allow any AI model to rifle through my personal data—especially not to create personalised, AI-generated memes based on my phone's photos. My digital memories are worth more than that—or, at the very least, they're worth an authentically terrible Photoshop attempt.

Some big changes out of EVE Online developer CCP Games today, including that it is no longer called CCP Games: The studio has parted ways with Crimson Desert maker Pearl Abyss, entered into an AI research partnership with Google DeepMind, and ditched its nearly 30-year-old name, rebranding itself as Fenris Creations.

Pearl Abyss acquired CCP—sorry, Fenris Creations—in 2018, and frankly I'm not sure what actually came out of that deal in terms of practical results. But after "a joint review of long-term strategy," which included an analysis of "differences in operating context, current strategic focus, and long-term priorities," the companies decided—after settling on a price tag of $120 million—to go their own ways.

Fenris said the change will only impact its ownership and governance, and that it will continue to operate as it always has, with no changes in personnel or development plans.

"The teams building EVE Online, EVE Frontier, EVE Vanguard, and EVE Galaxy Conquest remain in place, and our studios in Reykjavík, London, and Shanghai continue as they are today," still-CEO Hilmar Veigar Pétursson said in a message to the EVE community. "Our leadership, creative direction, products, and development plans are also unchanged. The people who have been working on New Eden for many years are still here, still building, and still focused on the same goal: making EVE stronger for the long term."

It's weird to see that historical studio name gone, and it'll no doubt take me ages to get used to the new one. But in terms of far-reaching impact, the much bigger deal here is the new AI research partnership with Google. As part of that deal, which will see Google take a minority stake in CCP Fenris Creations, an offline version of EVE Online will serve as a new training ground for the DeepMind AI, enabling it to "test and evaluate models in a controlled setting."

CCP—sorry, sorry, Fenris Creations—hasn't been shy about embracing new technologies in the past, including AI, Earlier this year, for instance, it launched an "AI-powered assistance feature" for EVE Online, trained on more than 5.8 million messages posted in EVE's Rookie Help channel, to help ease new players into the notoriously dense game.

But aspirations for the DeepMind deal sound loftier: The partnership will "explore new gameplay experiences" enabled by AI technology, but will also focus on more esoteric subject matter, including "long-horizon planning, memory, and continual learning." Pétursson said the deal with Google is a good fit because "EVE is one of the few environments where questions about intelligence can be explored inside something that already behaves like a living world."

Google DeepMind co-founder and CEO Demis Hassabis shared similar thoughts, saying that games have "been at the heart of many of Google DeepMind’s breakthroughs—like Atari DQN, AlphaGo, AlphaStar and SIMA—because they’re the perfect training ground for developing and testing AI algorithms."

"I’ve known Hilmar for many years and long admired his work, and I’m thrilled to partner with him and the fantastic team at Fenris Creations to explore new gaming experiences and advance AI research safely inside a player-driven universe as amazingly complex as EVE Online."

Separately from all of this, things are apparently going swimmingly for EVE: Pétursson said 2025 produced some of the game's "strongest results in years, including a record-breaking November and one of the strongest quarters in EVE Online’s more than 20-year history." CCP—ah jeez, you know who I'm talking about—remains profitable and has amassed "strong reserves" so it can continue investing in the future of EVE. Speaking of which, more on that, including a closer look at the research that will be conducted via the DeepMind partnership, will be shared at EVE Fanfest 2026, which kicks off on May 14.

My complaints about the generative AI bubble, and the tech industry in general, can often be reduced to the question 'What are we even doing here?'

Case in point, one of the things Google thinks will save the games industry is the ability to predict when someone is about to stop playing a game so that they can be manipulated, in ways invisible to them, into playing more. To which I have to wonder: What are we even doing here?

The concept was mentioned—not quite in those terms—by Google Cloud gaming exec Jack Buser back at the Game Developers Conference in March. His talk broadly promoted generative AI adoption, hopping across everyday game dev problems AI is being used for: tagging assets, debugging, detecting cheaters. I don't have an especially visceral reaction to those uses for AI, but my stomach churned at the mention of "hyperpersonalization."

AI is great! Here's how we're going to use it against you

AI boosters have been kicking the "hyperpersonalization" term around for a few years now. Microsoft said in a 2024 blog post that the outcome of hyperpersonalized AI marketing should be that "customers' needs are anticipated before they even ask."

When it comes to games, Buser is excited by the idea of generative AI models that invisibly adjust a player's experience so that marketing "feels like it's part of the gameplay experience."

"AI is doing this so well now that it can predict churn before it even happens in extremely robust ways," Buser said, "and it can adjust gameplay as well as offers that you put in front of your players in near real time."

If you're building software that makes Minority Report-style predictions about the likelihood someone will stop playing your videogame so that you can then manipulate them into playing and spending more, I think you should go sit by a body of water, listen to the birds, and ask yourself what the hell you're even doing anymore.

If it's truly impossible to make money selling games without using machine learning to build psychological prisons—like in that one Star Trek: TNG episode where everyone got addicted to willing discs into tubes—then there's no games industry to save. There are just software companies competing to design the best digital nicotine.

We might already describe parts of the industry that way—loot boxes, daily quests, and other tricks all predate AI—but certainly not all of it. Big and small developers are still making games that, whether or not they pay the bills, exist because their creators genuinely want to enrich our lives.

"Many people make games for money, but we make money for games," one triple-A studio founder recently told us. Balatro, despite being a certified time-devourer and one of the best-selling PC games in recent memory, came from a developer who dislikes gambling and says microtransactions make him "want to put [his] computer in the dishwasher." (He hates AI art, too).

But others, especially in Silicon Valley, have clearly decided that making numbers go up is all that matters, and expect us to be thankful that we've been deemed suitable targets for their next number extraction endeavor (or propaganda project). They tell us openly that their goal is to make more money while employing fewer people, and then act surprised when we don't applaud them for it.

Perhaps one day they'll be able to do away with the abstract idea of 'making a good videogame' altogether, designing hits purely by using AI to respond to the dictates of the numbers. They might end up with something like this self-propelled slot machine.

Jack Buser is Google Cloud global director for games, a role in which he "enable[s] transformation of game industry developers, publishers and platforms with AI and cloud solutions." And from that lofty perch he has devised a solution for the woes plaguing the videogame business. Can you guess what it is? I'll tell you: It's AI.

In an interview with GamesIndustry, Buser said the videogame business is "finally returning to revenue growth," but at the same time profits are declining, games are being cancelled, and layoffs continue to sweep the industry. The only real growth is coming from Roblox and China, and if you're not in one of those categories, "odds are you're struggling to some extent."

"Once you start to look underneath the surface of what's going on in the industry, you realize like, oh my gosh, we are in trouble," Buser said, referencing the decimation of the game industry that kicked into high gear in 2023 and hasn't eased up since.

"This is not a sustainable business model. We have to transform as an industry to meet this moment so that we can drive into the future. And so that frames up why this is such an important time in the industry."

Thus, AI: Buser said aspects of the business including marketing, business strategy, and analytics are being "accelerated radically with AI," and he also believes that embracing AI will reduce development times and costs, and enable smaller studios to compete more directly with the major players.

"A general trend we're seeing is the very large game companies are thinking about their development pipelines, reducing iteration time," Buser said. "But we're seeing the long tail, as well as the sort of torso of the industry start to realize that with AI, they can punch way above their weight, way above their weight, and they can actually compete with some of these larger budget games by leveraging AI."

Buser said AI is rather like Iron Man's suit—"put it on and see what types of superpowers it's able to grant you"—but he gives up the game a little bit near the end of the report. After saying the game industry is "broken," which is why we've seen tens of thousands of layoffs over the past few years, he explicitly positions AI as the solution to the problem—for the business. For the people who work in it? Well, not so much. But that's not the fault of AI!

"For somebody who's in a studio worried about their job, the existential nature of the games industry right now is putting a lot of pressure on these studios, and that is outside of anything to do with AI," Buser said.

"If anything, AI is going to help us right-size these business models. It's going to help us create a healthier industry, not just for the big players, but for small players as well. And I think if we have a more nuanced understanding of how it's actually being used and refine our language a little bit about how we discuss these things, it's easily one of the most important things we can do as an industry right now."

The giveaway is of course Buser's use of "right-size," a euphemism for laying off even more people—defined by the Cambridge dictionary as "the process of making a company or organization a more effective size, especially by reducing the number of people working for it." The goal, in other words, is not to protect people, but to protect profits—understandable from the perspective of a Google executive, I suppose, but not a solution I have much interest in.

Buser's previous gig was "global director for all games business development activities for Stadia," and we all know how that turned out: An overhyped promise of a bold, accessible future for gaming that never went anywhere except, eventually, away. It's not hard to see parallels with the relentless push for AI-driven game development, which we're repeatedly assured is an inevitable future: Yes, it may be useful for specific tasks like software debugging, but it has yet to come close to delivering on the hype. (I'm still waiting for an update on Elon Musk's promise to release "a great AI-generated game" by the end of 2026.)

I do agree with Buser that the conversation would be helped immensely by using better, more precise language in place of the catch-all of "AI": Technological advancement has always been an essential part of game dev, and that that's going to continue to be the case moving forward. He might even be correct in his broader thesis, to the extent that maybe the videogame industry as a whole can start making more money with less effort by leaning heavily into AI garbo: After all, if there's one thing we've learned from the advent of reality television, it's that creative bankruptcy will not prevent you from making serious bank. Sure, we all end up playing Courtroom Chaos Starring Snoop Dogg, and that'll suck, but at least the money guys will be alright.

Odds are that you didn't just stumble across this story via the PC Gamer homepage—it's way more likely that Google had something to do with it. Maybe Google Discover served you this story, or maybe you dove into Google search in a bid to collate all of the news you can find about various government bodies attempting to challenge the search giant's crown. The European Commission is the latest to propose such measures.

Specifically, the European Commission is proposing that Google share its search data with third party search engines. The data would include "ranking, query, click and view data" and ideally be presented "on fair, reasonable and non-discriminatory terms" according to the European Commission.

The hope is that by sharing this data, "third party online search engines, or ‘data beneficiaries', [would be able to] optimise their search services and contest Google Search's position."

These proposed measures are in a bid to comply with the Digital Markets Act, which came into effect throughout the European Union in late 2022. The act targets 'gatekeepers,' or "large digital platforms providing so-called core platform services, such as online search engines, app stores, [and] messenger services," in an attempt to regulate the multimillion dollar companies that now shape how we interact with one another and the world around us.

Teresa Ribera, the Executive Vice-President for Clean, Just and Competitive Transition, explained, "Data is a key input for online search and for developing new services, including AI.

"Access to this data should not be restricted in ways that could harm competition. In fast-moving markets, small changes can quickly have a big impact. We will not allow practices that risk closing markets or limiting choice."

This isn't the first time the sharing of Google's search data with its rivals has been proposed. You'll remember that a US judge ruled that Google had violated antitrust laws back in 2024, resulting in a number of proposed remedies. These included the suggested divestment of the Chrome browser, though ultimately that did not come to pass.

In September 2025, the US Department of Justice also ordered Google to "make certain search index and user-interaction data available to rivals and potential rivals." Vice president of regulatory affairs at Google, Lee-Anne Mulholland, claimed that data sharing would "risk Americans’ privacy and discourage competitors from building their own products—ultimately stifling the innovation that keeps the US at the forefront of global technology."

So, in January of this year, Google requested to delay data sharing while it appealed the 2024 antitrust ruling.

In other words, Google has shown a lot of resistance to data sharing proposals in the past. So, what does that mean for the European Commission's latest move? That remains to be seen, as both interested parties, as well as Google, have until May 1 to submit their views on the proposed measure.

Taking all of that feedback into account, the European commission will then come to a final, binding decision that Google must then adopt by July 27. Whether that decision knocks the corporation from its search pedestal, or only gives Google a wee nudge, time can only tell.

If you click on almost any major website, you'll surely have notice that you get asked about cookies. These files store your preferences and help companies track your data. However, it turns out that even if you decline to use them, there's a good chance you're still being tracked.

A recent audit from webXray has found that "major technology companies simply ignore globally defined opt-out signals" (via TechSpot). The report states that it observed 194 online advertising services ignore standard opt-out services, plus cookie banners certified by Google that reportedly "fail to prevent Google from setting cookies after users opt out with a globally standard signal."

webXray studied 242 ad tech vendors in total, which means an 80% failure rate to adequately opt-out users, according to the report.

The analysis shows that 55% of sites set ad cookies despite users opting out, and 78% of cookie banners failed to protect their users. It argues that companies' liability exposure racks up to a whopping $5.8 Billion.

webXray has examined Google's own cookie system and reckons that when it sends an encoded cookie, it then reportedly responds by creating a new advertising cookie named IDE. The report says, "This non-compliance is easy to spot, hiding in plain sight," and makes the case that Google should instead respond to encrypted cookies with a 451 code, stating "unavailable for legal reasons." webXray states that Google failed to adequately opt-out users in 86% of cookies, with over 11,000 cookies set, despite user preferences.

Microsoft and Meta are also highlighted in the report. It's claimed that both companies fail to adequately respect cookie opt-out requests. It states Microsoft creates a new MUID cookie, even when sent an encoded one, and reports it failing to opt-out users in 50% of attempts. This means that over 7,500 cookies were sent, despite opting out.

The report argues Meta's Pixel tracking code fails to check for opt-out signals, with it getting a 59% opt-out failure rate in testing. That works out to over 1,200 cookies set despite opting out.

webXray concludes that cookies are now a "legal minefield that puts users at risk", and one can hope the worry of legal fees and other penalties will eventually correct the problem.

After several years of uncertainty, job cuts, issues with self-destroying CPUs, and spending billions of dollars on its foundries, Intel has made big strides in moving into a better place. Thanks largely to its latest products and deals with other tech and AI giants. Investors certainly seem to think so, because its market cap is now the highest it has been in more than two and a half decades.

As reported by Tom's Hardware, you can see the figures yourself and easily spot that the last time Intel's market capitalization was above the $300 billion mark was all the way back in October 2000. According to CompaniesMarketCap, the site tracking the figures, this makes Team Blue the 47th most valuable company in the world, though just by market cap and nothing else.

In comparison, AMD is currently enjoying a cap of $385 billion, and Nvidia is miles ahead of either company, with a market cap that's approximately $4.5 trillion in size. However, this is all somewhat of an unfair comparison, because neither of those firms have their own manufacturing plants for CPUs and GPUs, whereas Intel does.

You'd think that this would result in a higher market cap, but for the majority of Intel's history, it's only ever made chips for itself, and those fabs have been a serious financial drain in the past. But now that it's agreed to be part of Elon Musk's Terafab project, there's a chance that Intel's numerous foundries could be busy creating or packaging AI chips for SpaceX and Tesla.

Intel has already recently agreed to a chip deal with Google, and along with how well its Panther Lake and Arrow Lake 200S Plus chips have been received, it would seem to suggest that investors believe that the 57-year-old company is on the right track to regaining its former glory.

It's certainly faring better than other US-based chip companies, such as Texas Instruments and Qualcomm, which are either relatively static or currently on a bit of a slide, in terms of market cap.

Chasing down manufacturing orders seems to be what investors are most interested in Intel doing right now, probably because no matter what its desktop CPUs are like (or what they do to themselves), Intel's overall processor market share is still pretty dominant. AMD is certainly making headway in the world of server chips, though, and when it comes to PC gaming and consoles, Team Red rules the roost.

A healthy AMD and Intel is good news for consumers, because having just one company utterly dominate a market isn't great for competition (*cough graphics cards *cough). Investors are perhaps less concerned by this, but they're certainly happy with how things are going at Intel right now.

In a sublime example of the internet's "I made this" effect, an Italian TV channel temporarily took down Nvidia's March 16 video "Announcing Nvidia DLSS 5" via copyright strike, claiming the content as its own.

As reported by NikTek on X.com, the DLSS 5 reveal trailer was unavailable in Italy for more than 24 hours, with the message "This video contains content from La7, who has blocked it in your country on copyright grounds" taking its place. While we're doing a little reverse-engineering here to understand what happened, La7 seemingly used footage from DLSS 5 in a recent broadcast, then started issuing copyright claims based on that footage.

Brazen? Yes. Funny? Even more so, considering the overt distaste both players and the games industry have shown for DLSS5.

The impact extended further than Nvidia's own YouTube channel. Multiple other posters on X chimed in to share their own videos being demonetized and region blocked due to a copyright claim. Former IGN editor Destin Legarie posted about the incident on April 4, prompting a rather unhelpful response from YouTube's official X account explaining the basics of its Content ID system.

"The channel making these claims does not own the rights to an Nvidia trailer they used without permission during their talk show," Legarie shot back. "They just created work for YouTube by abusing the system."

Legarie posted on Monday that La7 has released the claim. Using a VPN, PC Gamer confirmed that Nvidia's reveal trailer is also now once again viewable in Italy.

YouTube's Content ID system is often criticized for being abusable, with frivolous claims on videos that should fall under fair use often meaning smaller channels see their videos demonetized unless they can contest a claim, a process that can take weeks. It's equally common for these automated systems to go too far, ending up in claims that are quickly revoked as soon as a real human being takes notice.

It's funnier to think of someone at TV channel La7 deciding it owns all coverage of Nvidia DLSS 5 after producing one talk show segment about it, but it seems more likely that someone clicked a "protect our copyright" button without fully understanding what it would entail.

Or maybe La7 was gung ho about taking credit for the new technology until it realized it was picking a fight with Dave Oshry, at which point the only safe move was immediate surrender.

Thanks to the breathtaking power of artificial intelligence, we may now use natural language to beg the computer to perform a function that once came as naturally to its binary brain as breathing does to ours. Google—a company worth 3.5 trillion dollars, uncontested owner of both the almighty internet search algorithm and the web browser, has overengineered its latest Discover feed AI curation tool to such an extent that it cannot simply block a website from appearing as a source.

Google's 'Tailor your feed' feature, introduced a few months ago, adds a familiar chat prompt that encourages you to type something like "Keep me updated on what's happening in college basketball" to curate what shows up in your Discover feed. Discover, if you're not familiar, is what Google calls the list of recommended stories that shows up when you open a new tab in Chrome, and is also built into the Google app and Android.

'Tailor your feed' is an experimental "Lab" feature you have to enable manually. I can't recommend it, because it will make you feel mad, sad, and dumb, not necessarily in that order.

I'm no computer scientist, but my understanding is that all the stuff we're doing On Here ultimately distills down to, y'know, binary: on or off, yes or no, zero or one. On one hand, this is such an oversimplification of technology that it may have little bearing on the complexities of the large language models and mystery algorithms that power modern AI agents. On the other hand, I don't think that actually matters when the end result of that modern complexity is failing at a task as simple as yes or no. I enabled 'Tailor your feed' so that I could tell Discover to completely block X.com links from appearing in the list of curated stories, because I really don't need to see what the worst people in the world are saying when I'm scrolling for headlines about new brunch spots in my neighborhood. And here's what happened when I made that request:

You won't see as many posts from X going forward.

I can't guarantee you'll see none of something.

As I found with a quick search on Reddit, I'm not the only person who's tried this solution since Google started flooding Discover with posts from X in late 2025. (Given that Google could pull quality writing from a billion websites and blogs across the internet, I have to wonder why it instead decided the site that was plagued with AI-generated child sexual assault imagery was an ideal source.) After typing in something similar, one Redditor reported seeing only "one story from X as opposed to the usual dozen."

"It needs constant reminding but better than it was…" the Redditor wrote recently.

So is Google placing its thumb on the scale by refusing to remove a paying partner entirely from its feeds, or has it developed a piece of AI software so staggeringly stupid that it can't even create a blacklist (or 'blocklist' as they're more often called these days)? The entire damn internet is built around blocklists. Google Chrome has built-in blocklists of websites it deems unsafe. Every piece of antivirus software in existence? Blocklists. Internet Explorer had blocklists in 1996!

What are we doing here, man. "I can't guarantee you'll see none of something" should be a vastly more difficult operation for a computer than "zero." Google Labs is out here spending billions of dollars trying to unsolve problems. This new frontier for software is indeed revolutionary, because I've never simultaneously felt embarrassed for the computer and for myself when trying to coax it into the simplest possible action.

It's clearer than ever that big tech's vision for the algorithms choking the life from the internet springs from the idea that they know better than we do what we actually want. It really is time to turn off the recommendations and ditch this stuff once and for all. RSS awaits.

Google published a whitepaper yesterday that dramatically pulled in the horizon on bitcoin and other cryptocurrencies. Google Quantum AI researchers claim that cracking the cryptography protecting many major coins, including bitcoin and ether, could require a quantum computer with fewer than 500,000 physical qubits, something in the order of 20 times lower than previous estimates, Forbes reports.

Indeed, one of the paper's co-authors, Justin Drake, has concluded that the timeframe for "Q-Day", the hypothetical moment when a sufficiently powerful quantum computer becomes capable of breaking the public-key encryption algorithms that underpin much of modern digital security including cryptocurrencies, could arrive much sooner than previously thought.

"My confidence in Q-Day by 2032 has shot up significantly. IMO there's at least a 10% chance that by 2032 a quantum computer recovers a secp256k1 ECDSA private key from an exposed public key," Drake said on X.

To put all this into at least some kind of context, Google's current most powerful quantum device, the Willow chip, has 105 qubits, while IBM's Condor processor was the first superconducting qubit processor to surpass 1,000 qubits, notching up 1,121 qubits.

However, the simple qubit count can be a somewhat misleading metric. Google's 105-qubit Willow is in many ways more capable than systems with far more qubits, because it achieves fidelities of 99.97% for single-qubit gates, 99.88% for entangling gates, and 99.5% for readout across its entire 105-qubit array.

According to Google, a computer with 1,000 low-quality, error-prone qubits can easily be outperformed by one with 100 highly reliable ones. As a consequence, the quantum computing industry is shifting to measuring logical qubits or error-corrected groupings of physical qubits as a more meaningful measure.

It's also important to understand that Q-Day doesn't mean that all digital security is immediately toast. Broadly, there are two kinds of cryptography. Public-key cryptography relies on mathematical problems (factoring huge numbers, discrete logarithms and so on) that quantum computers can solve via Shor's algorithm. This is what Q-Day would actually represent.

Then there's symmetric encryption and hashing, including AES and SHA-2. Those are vulnerable to Grover's algorithm, which effectively halves the security strength rather than dismantling it entirely. In that scenario, for instance, AES-256 encryption would be degraded to AES-128 levels of security. What's more, using longer keys in the first place would restore protection.

All of which means that Q-day would be less an instant end to all encryption and more a case of game over for public-key cryptography while everything else gets a little weaker and faces a painful but plausible migration to longer keys.

As Forbes reports, "Google’s whitepaper points out that cryptocurrencies are uniquely exposed among systems that rely on this type of cryptography. Blockchains use elliptic curve keys that are almost an order of magnitude smaller than RSA keys at comparable security levels, meaning a smaller quantum computer can crack them.

"And unlike traditional finance, which layers multiple safeguards, blockchains offer no recourse against fraudulent transactions. One forged signature could mean irreversible theft."

Indeed, Google's new paper, which is titled "Securing Elliptic Curve Cryptocurrencies against Quantum Vulnerabilities," does rather put a different spin on the company's declaration, which we reported on recently, that it is effectively prepping for the "quantum apocalypse" and planning to move over to post-quantum security measures by 2029.

Of course, for now this is just a prediction. Nobody has yet built a quantum machine with anything close to even the substantially reduced qubit count mooted in the new Google paper. But Q-Day does seem to be the opposite of many future technologies like fusion energy and a cure for cancer.

Instead of remaining slightly out of reach, always roughly the same way off into the future, Q-Day and the presumed demise of bitcoin seems to accelerating towards us.

One of the most popular JavaScript libraries, Axios, was recently the victim of an attack that had fake, malicious versions available to roll out to developers. These malicious versions install a remote access trojan (RAT), which is, as the name implies, a kind of malware that allows an attacker to access compromised devices from a remote location.

Google has identified the attackers responsible as likely being UNC1069, "a financially motivated North Korea-nexus threat actor" that goes by CryptoCore.

They compromised the Axios maintainer's npm account, npm being a trusted online registry of JavaScript code for users to share and use. Two poisoned packages were added to the Axios npm, and these added a new dependency that installs a RAT.

Malicious code never got into the official Axios software itself, which remains safe, but instead two separate malicious versions were published from an account that usually publishes legitimate Axios versions. Given the way npm works, these compromised, fake versions were able to be pushed to some developers.

The attack was staged almost a day in advance, the two poisoning attacks were timed pretty precisely, and evidence was erased post-exploit, pointing towards a calculated rather than opportunistic attack.

As cybersecurity company StepSecurity explains: "This was not opportunistic. It was precision. The malicious dependency was staged 18 hours in advance. Three payloads were pre-built for three operating systems. Both release branches were poisoned within 39 minutes of each other.

Every artifact was designed to self-destruct. Within two seconds of npm install, the malware was already calling home to the attacker's server before npm had even finished resolving dependencies. This is among the most operationally sophisticated supply chain attacks ever documented against a top-10 npm package."

However, it's important to note that developers using Axios wouldn't have been automatically infected. The malicious versions would have been automatically installed by many projects whenever they next run an npm install command. How often this command is run depends entirely on the company—maybe every week or two, or with a new package install.

Given that the malicious versions were removed within a few hours, it's likely that most developers using Axios are safe. However, BitDefender says its "telemetry confirms RAT execution attempts on customer systems, blocked by GravityZone and says "the blast radius is not theoretical."

The company recommends identifying exposure, assessing for prior compromise, and monitoring outgoing. Malwarebytes says: "If you are a developer deploying Axios, treat any machine that installed the bad versions as potentially fully compromised and rotate secrets. The attacker may have obtained repo access, signing keys, API keys, or other secrets that can be used to backdoor future releases or attack your backend and users."

Someone from a cybersecurity site and educational malware repo, VX-Underground, recently explained the severity of this on X as follows: "The impact from Axios being compromised is devastating, the fallout from this will be a massive headache. This is unironically a malware nuclear missile and will likely be studied in the future."

Google's TurboQuant AI compression algorithm caused quite the stir upon its announcement last week, as it promised to cut AI model memory usage by around 6x. The news of the vector compression-enhancing algo seemed to coincide with a drop in stock price of the big three memory chip suppliers, although other market factors like production material supply shortage concerns caused by the ongoing conflict in Iran likely took their toll, too.

Still, the market shake up is currently suggested to be behind a drop in price for consumer RAM listings. Amid reports of plummeting DDR5 prices, I decided to take a look at the pricing history of some of the most popular RAM kits on the market, and yep, some of them are going down alright—but I think it's a little early to be celebrating just yet.

Take this Corsair Vengeance RGB DDR5-6000 32 GB RAM kit, for example. It's the very same set of sticks I use in my personal machine, and one I bought for around $90 three years ago. That tracks with the data provided by Camelcamelcamel, where the kit trucked along at around the $90 mark until November of last year, until it soared into the $400+ range, topping out at $440 or so.

The kit wavered around this price until early March, where it dipped to $370, and appears to have stayed there ever since. That's a significant decrease from its earlier highs, although it dropped well before the TurboQuant announcement. Still, savings are savings—although it's by no means down to a price that I'd call reasonable.

Corsair's DDR5 offerings seem to be significantly cheaper than most other memory modules at the moment, which looks to be the driving factor behind some of these reports. Looking at other manufacturer prices, though, the picture becomes less rosy.

While this Crucial Pro DDR5-6400 32 GB kit is currently down to $385 at Amazon, a $35 saving over its $420 highs around the New Year period, the pricing data shows it hovering in the mid-$350 range in January, before jumping to $400 at Camelcamelcamel's last count.

Which does expose a slight flaw in the methodology here. Camelcamelcamel is one of the most respected price trackers on the interwebs, but it's not always the most up-to-date with its information. So instead, I took an anecdotal look at DDR5 prices across various different retailers to see if I could find significant drops elsewhere in the consumer RAM market.

And what I found were discounts. Now, as someone who puts together the best gaming PC build guide for both this website and PC Gamer magazine every month, this is something of a surprise. I've had my head in my hands since last November trying to find anything approaching a deal on decent DDR5, so a slew of "save!" tags looks suspiciously like a cool glass of water in an ever-drying desert.

They're not exactly mega savings, though. This G.Skill Trident Z5 32 GB DDR5-6400 kit has $50 off at Newegg right now, which would normally be something to shout about—except it's been knocked down from $550 to $500. An absolute bargain, I think not.

Or there's $60 knocked off this Team Group T-Create Expert 32 GB DDR-6000 kit, pushing it down to… $440. Or a $50 saving on a Crucial Pro OC 64 GB kit, which means it's still an eye-watering $630. It's lucky our Nick has discovered that 16 GB of RAM is still enough for gaming in 2026, isn't it?

Still, it's pleasing to see RAM discounts back on the menu, no matter how small they might be in the grand scheme of things. But a market-wide shift towards significantly lower prices, or a straight up plummet down to previous levels, doesn't seem to be on the cards just yet.

It remains to be seen whether TurboQuant, a sluggish PC part market, and wavering DRAM manufacturer stock prices end up having the desired effect. In all honesty, it's probably far too early to tell. But as things stand right now, it's still looking like an awful, awful time to buy RAM for your gaming PC.

Put the champagne on hold: I think it's likely to be a while yet before we can celebrate the return of reasonably-priced RAM.

Other than the AI bubble bursting or hype dying down, the other thing that could allow the RAMpocalypse to ease off is a technological change that leads to a dramatic reduction in how much memory AI needs. To that end, Google has cooked up TurboQuant, a new compression algorithm that promises to reduce memory demand by about 6x. And memory maker stock prices have already dropped, likely as a result of this.

Although we should resist being reductive and assuming Google's new algorithm is responsible for these market changes—lest we forget the effects on crucial material availability thanks to the war in Iran—a 6x claimed reduction in memory demand must surely account for at least some of it.

According to Google, TurboQuant "optimally addresses the challenge of memory overhead in vector quantization" and "achieves a high reduction in model size with zero accuracy loss."

In other words, it makes vector compression—which is critical for AI models understanding and processing information, as they do so using vectors—require less memory than it has until now and crucially without the normally associated loss of accuracy from compressing things down.

The basic idea, to exclude lots of details and simplify greatly, seems to be a shift from calculating things in terms of standard vectors and instead using a more absolute reference system. Which, to my non-mathematical ears at least, sounds a bit like moving away from vectors:

"Instead of looking at a memory vector using standard coordinates (i.e., X, Y, Z) that indicate the distance along each axis, PolarQuant converts the vector into polar coordinates using a Cartesian coordinate system. This is comparable to replacing 'Go 3 blocks East, 4 blocks North' with 'Go 5 blocks total at a 37-degree angle'."

This, ultimately, means no need for data normalisation, which should "eliminate the memory overhead that traditional methods must carry." Google has put the new algorithm through its paces in a bunch of benchmarks, and the results, according to the Big G, at least, show that "TurboQuant achieves perfect downstream results across all benchmarks while reducing the key value memory size by a factor of at least 6x."

Again according to Google, the results also "demonstrate a transformative shift in high-dimensional search... [allowing] for building and querying large vector indices with minimal memory, near-zero preprocessing time, and state-of-the-art accuracy."

Naturally, such a big improvement, if true, could drastically change the AI server market. Which means it could change the amount of memory that AI companies are wanting to buy from Micron, SK Hynix, and Samsung.

As identified by investor boffins, that is indeed what the market is predicting we will see, because stock prices for the big three memory makers have already dropped. Samsung's, for instance, has dropped by about 8% since a couple of days ago, SK Hynix's by about 11%, and Micron's by about 10%. And these are all actually after factoring in a slight rebound today.

If AI companies need to buy less memory, that will of course raise the amount of supply open to the general consumer market, including for gaming PCs, laptops, handhelds, and other goodies. Which should in theory mean memory gets cheaper.

This highlights the stark difference between what's good for memory makers and what's good for we end users. It's something I've noticed a lot over the past few months: the less stock and the more demand there is, the happier memory investors and analysts seem to be, and the unhappier consumers are. Which is obvious, of course, but it's interesting to see it to go the other way for a change.

We shouldn't consider anything a done deal, though. After all, Micron has already said there is "demand significantly in excess of our available supply for the foreseeable future." So for all we know, much of the 'freed up' memory production could just go straight back into AI server racks rather than our PCs. There's also the fact that compute hungry AI folk will just end up running larger models if the memory requirement drop. But there's a chance, at least, and I'll keep clutching at it.

If you've never heard of Q-Day, the moment when quantum computers become capable of breaking traditional encryption methods and exposing vast amounts of data, often referred to as the 'quantum apocalypse', then Google's latest announcement might come as something of a shock.

The company is giving itself a 2029 timeline to "secure the quantum era" by migrating over to post-quantum cryptography (PQC) methods before the old ones are broken by existing quantum computers. And being one of the primary players in the space, Google's hoping that if it starts making major changes now, the rest of the tech world might follow suit.

"As a pioneer in both quantum and PQC, it’s our responsibility to lead by example and share an ambitious timeline", the announcement reads. "By doing this, we hope to provide the clarity and urgency needed to accelerate digital transitions not only for Google, but also across the industry."

"Quantum computers will pose a significant threat to current cryptographic standards, and specifically to encryption and digital signatures", says the company. "The threat to encryption is relevant today with store-now-decrypt-later attacks, while digital signatures are a future threat that requires the transition to PQC prior to a Cryptographically Relevant Quantum Computer (CRQC)."

Google says that it's adjusted its threat model to prioritise this migration for its authentication services, and it recommends other engineering teams "follow suit".

Google has often claimed significant (and debateable) advances in its quantum computer development over the years, so perhaps there's a touch of humble brag to this timeline, too. I've got a scary-powerful quantum computer and it's only getting better, that sort of thing.

Still, the threat of traditional cryptographic keys being broken in one quantum sweep appears to be a legitimate one, as evidenced by the preparation of organisations like the US National Security Agency (NSA). Until recently, however, the running joke was that Q-Day was always at least a decade away, a bit like viable nuclear fusion or Half Life 3. I know, I shouldn't have mentioned it.

This new timeline, though, would suggest that Google is earnestly concerned about the potential impacts of such an event and wants the world to speed up in its ongoing efforts to get ahead of it. Personally, I'm still trying (and failing) to wrap my brain around exactly how the company's quantum computing tech works, never mind the wide-ranging implications it might have for security measures going forward.

Anyway, the clever people need to hurry up, or security stuff gets all broken. That, I can get my head around. Thanks, Google. It's a brand new worry for my brain to gnaw on while I try to fall asleep tonight.

Late last month, Anthropic refused to remove AI autonomous weapons and mass surveillance safeguards when the US government asked it to do so. The administration then cancelled its contract in response. OpenAI took up that contract, and the Department of War soon threatened to label Anthropic a 'supply chain risk'. In yet another twist and turn in the story, Anthropic is suing the US government in response.

According to Anthropic CEO Dari Amodei, "They have threatened to remove us from their systems if we maintain these safeguards; they have also threatened to designate us a “supply chain risk”—a label reserved for US adversaries, never before applied to an American company—and to invoke the Defense Production Act to force the safeguards’ removal. These latter two threats are inherently contradictory: one labels us a security risk; the other labels Claude as essential to national security."

Anthropic filed its complaint (PDF warning) on March 09, claiming that Anthropic held to its principles of building AI that "maximises positive outcomes for humanity" and the US government, in turn, "retaliated against it for expressing that principle."

In response to the initial claim from Antropic, Under Secretary of War Emil Michael accused Antropic CEO Amodei of being a 'liar' and having a 'god complex'. Michael argues, "He [Amodei] wants nothing more than to try to personally control the US Military and is ok putting our nation’s safety at risk."

Notably, the suit mentions that Anthropic agreed to much of the Department of War's request to discard usage restrictions, except for two, which are for the use of the technology in lethal autonomous warfare and mass surveillance of Americans. It goes on: "Throughout these discussions, Anthropic expressed its strongly held views about the limitations of its AI services. It also made clear that, if an arrangement acceptable to the Department could not be reached, Anthropic would collaborate with the Department on an orderly transition to another AI provider willing to meet its demands."

Supporting its case, OpenAI and Google have both signed (PDF warning) an amicus brief (where a party with an interest in the case, who is not directly involved, gives some sort of testimony or argument) in support of Anthropic. It is likely in the best interest of OpenAI, which is currently working with the US Department of War, and Google, which has invested heavily in AI, to support any action made to stop the US government from blacklisting AI companies.

The brief declares, "Defendants recklessly invoked national security authorities intended to protect the procurement process from interference by foreign adversaries. If allowed to proceed, this effort to punish one of the leading U.S. AI companies will undoubtedly have consequences for the United States’ industrial and scientific competitiveness in the field of artificial intelligence and beyond. And it will chill open deliberation in our field about the risks and benefits of today’s AI systems."

OpenAI tagging in here is interesting, given CEO Sam Altman admitted to rushing out a deal with the Department of War and a hardware leader resigning in response. OpenAI has taken criticism for its deal both externally and internally, and has been in a bit of a weird place ever since.

Luiza Jarovsky, the cofounder of the AI, Tech and Privacy academy, who holds a PhD in Law, reckons "Anthropic is likely to win" and that it could lead to a precedent in regard to AI governance in the US. Getting the support of both Google and OpenAI only bolsters its defence.

Anthropic is seeking a hearing to declare the Department of War's action a violation of the First and Fifth Amendments of the United States Constitution and to seek damages, plus payment of any attorney fees. It also pleads that the judge grant "such further and other relief as this Court deems just and proper." A hearing could be held as soon as this Friday.

Earlier this week, the Google Threat Intelligence Group (GTIG) published a report on an exploit kit specifically targeting older Apple iPhones. Those with an up-to-date iOS may only feel momentarily smug as it turns out the kit, called Coruna, can sink its hooks into a wide range of phones—though the malicious range of this exploit kit is far from the worst wrinkle in this story.

GTIG says it tracked Coruna's use throughout 2025, beginning with "highly targeted operations initially conducted by a customer of a surveillance vendor." However, the exploit framework is unlikely to have been built by cybercriminals alone, and may originate from hacking tools used by the US government.

Device security company iVerify has recently issued its own report on what it's calling the 'First Known Mass iOS Attack,' claiming that the exploit chain at the centre of Coruna "has similarities to previous frameworks developed by threat actors affiliated with the US government."

"While iVerify has some evidence that this tool is a leaked US government framework, that shouldn’t overshadow the knowledge that these tools will find their way into the wild and will be used unscrupulously by bad actors", says the researchers.

GTIG says Coruna consists of "five full iOS exploit chains and a total of 23 exploits," two of which bear a striking resemblance to iOS exploits "that were also used as zero-days as part of Operation Triangulation." Triangulation was a 2023 hacking operation targeting Russian cybersecurity firm Kaspersky. The Russian government alleged the NSA was behind it, though the US government has neither confirmed nor denied this.

The full Coruna framework can be levelled at iPhone models running iOS version 13.0 (released in September 2019), all the way up to version 17.2.1 (released in December 2023). Coruna can quietly infect what is potentially a lot of phones, and then be used to harvest swathes of sensitive data (including photos and emails), as well as steal cryptocurrency.

GTIG was able to extract the full exploit kit from an attack by "UNC6691, a financially motivated threat actor operating from China." But the team additionally reported it also saw the exploit framework deployed in earlier attacks against Ukrainian users by suspected Russian threat actor UNC6353.

This is cause for concern all on its own as it suggests cybercriminals are trading tips on how to carry out malicious attacks internationally, and that there's "an active market for 'second hand' zero-day exploits." The alleged US government origin lore makes that all the more dreadful.

IVerify's report sums it up, saying, "Despite assurances from commercial spyware developers and the governments who purchase them that use will be limited to counterterrorism, only against criminals and by non-authoritarian administrations, the reality has begun to settle in: once spyware or an exploit capability is sold, control over the end customer is lost."

To put it another way, and to paraphrase Jensen Huang, that's why I'd argue manufacturers of consumer electronics like phones and PCs shouldn't offer up 'secret backdoors'—you can't guarantee who exactly will end up with the keys.

Google, Microsoft, Meta, Amazon and "several artificial intelligence companies" signed a pledge at the White House yesterday, dubbed the “Ratepayer Protection Pledge,” to pay for the cost of powering their data centres, the Guardian reports. But there are questions over the legal authority of the pledge and what it will amount to in practice.

The move reflects concerns in the US that the proliferation of data centres, driven largely by AI, is driving up the cost of electricity for home owners and businesses. "President Trump is calling on the leading United States hyperscalers and AI companies to build, bring, or buy all of the energy needed for building and operating data centers, paying the full cost of their energy and infrastructure, no matter what," the White House statement on the Pledge said.

“This means that the tech companies and the datacenters will be able to get the electricity they need, all without driving up electricity costs for consumers,” President Donald Trump said.

“This is a historic win for countless American families and we’ll also make our electricity grid stronger and more resilient than ever before.”

The pledge commits participating tech companies to pay for new or expanded power plant capacity and also to sign up to special rates for power supplies.

The pledge includes a commitment by technology companies to bring or buy electricity supplies for their datacenters, either from new power plants or existing plants with expanded output capacity. It also includes commitments from big tech to pay for upgrades to power delivery systems and to enter special electricity rate agreements with utilities.

Notably, this includes tech companies paying those rates even if they don't actually use the power. "This will protect the American people from increased utility bills as a result of the development of these data centers," the White House said.

Meta, for one, is bigging up its role in energy supply and use, saying, "We pay the full costs of our data centers’ energy and water use so they aren’t passed on to consumers, and fund new and upgraded infrastructure."

In terms of who the other signatories were at the White House event, Oracle, xAI and OpenAI are reported to have been in attendance, but the White House release didn't include a full list.

It's also not clear what legal authority the "Pledge" carries. In a "Proclamation" concerning the Pledge, the President says that the "Ratepayer Protection Pledge and the commitments embodied therein effectuate the national policy of the United States...by virtue of the authority vested in me by the Constitution and the laws of the United States."

However, the Proclamation cites no specific statute or constitutional provision as the source of that authority. As things stand, then, it's hard to say what mechanism could be used to enforce signatories to meet their Pledge commitments.

It's worth noting that a Presidential "Proclamation" is not the same as an Executive Order. The latter essentially instructs federal agencies, departments, and officials on how to implement policy or interpret law and has much clearer legal authority, though still bounded by the Constitution.

That said, and by way of example, Trump has used both Executive Orders and Proclamations to implement tariff policy. It's just that this particular Proclamation is particularly thin when it comes to laying out its legal and constitutional standing.

So, we'll have to wait and see if this amounts to genuine actions by the tech firms in question, or if perhaps this is more about political positioning as the US mid-terms approach later this year.

Cybersecurity provider Malwarebytes has thrown up a red flag regarding a fake Google Account security page that it says is distributing "what may be one of the most fully featured browser-based surveillance toolkits we have observed in the wild", and it's capable of infecting Windows, Apple, and Google Android devices.

In a blog post breaking down the methodology of the attack, Malwarebytes says that the infiltration begins with what appears to be a genuine Google Account security check, from a page with Google's familiar stylesheet and with an official-looking domain.

A prompt then asks to install "security software" via a Progressive Web App (PWA) as part of a four-step process, which proceeds to gradually grant the attacker access to notifications, contact lists, real-time GPS location, and the contents of the host machine's clipboard, among others.

If a victim installs the PWA and grants requested permissions to the site, simply closing the tab is not enough to prevent it from access. The page script itself runs as long as the app or tab is open, and attempts to read the clipboard, looking for "one-time passwords and cryptocurrency wallet addresses". It also attempts to intercept SMS verification codes on mobile devices, and polls the API every 30 seconds as it waits for operator commands.

However, with the app and tab closed, a separate service worker runs malicious, data-stealing tasks in the background, and even queues stolen data locally if the device goes offline, before sending its payload as soon as the connection is restored.

"Close the browser tab and the page script stops. Clipboard monitoring and SMS interception end immediately," says Malwarebytes. "But the service worker remains registered. If the victim granted notification permissions, the attacker can still wake it silently, push a new task, or trigger a data upload without reopening the app. And if the victim ever opens it again, collection resumes instantly."

The malware also operates as a WebSocket relay, which means it can act as an HTTP Proxy and be used to gain access to corporate networks, bypassing IP-based access controls and funnelling traffic through the victim's IP address. "Once connected, the attacker can route arbitrary web requests through the victim’s browser as if they were browsing from the victim’s own network", says Malwarebytes.

The fun doesn't stop there, either. On Android devices, a separately-installed APK disguised as a "critical security update" includes a custom keyboard capable of capturing keystrokes, a notification listener for capturing two-factor authentication codes, an accessibility service that can observe screen content, and an autofill intercepting service to capture user credential fill requests. Oh, and microphone recording, of course. Fabulous.

Phew. It's about as comprehensive as malware gets by the looks of things, and getting rid of it seems to be something of a convoluted process. Malwarebytes provides step-by-step removal instructions for Windows and macOS users, including Chrome, Firefox, and Safari-specific options, along with some Android and iOS-focused steps to take if you've been duped into falling for its charms on your mobile device.

It's certainly one of the most nefarious-looking trojans I've ever read about, and the fact that it's capable of gaining access to your system via most of the popular browsers on Windows, Apple, and Android devices is deeply concerning. Pay attention to what you're clicking on, folks—it's a dangerous world out there.

Google Sheets is perhaps my most disliked member of the Google Workplace suite. It's not that it's bad at what it does, more that it's a deathly-dull spreadsheet editor that I loathe having to stare at for more than five minutes.

But lo and indeed behold, because Google says it's caught Sheets being used in a super-exciting act of global espionage! Okay, exciting was the wrong word. Concerning, that's what I was going for.

According to Google's most recent Threat Intelligence blog post, last week the Google Threat Intelligence Group (GTIG), alongside its partners, "took action to disrupt a global espionage campaign targeting telecommunications and government organisations in dozens of nations across four continents."

The threat actor, mysteriously named "UNC2814" and said by Google to be suspected of connection to the People's Republic of China, was said to be using API calls to communicate with SaaS apps and "disguise their malicious traffic as benign."

And would you believe it, the primary SaaS app in question was none other than our old friend, Google Sheets. At this point, I'd like you to imagine me ripping a Scooby Doo-style mask off a spreadsheet.

The mechanism by which our alleged spies operated is referred to by Google as Gridtide, and is described as a "sophisticated C-based backdoor with the ability to execute arbitrary shell commands", as well as uploading and downloading files.

Gridtide is said to have been leveraging Google Sheets as "a high-availability C2 platform, treating [a] spreadsheet not as a document, but as a communication channel to facilitate the transfer of raw data and shell commands."

If anyone else is thinking of poor Google Sheets being marched at gunpoint past security and into a bank vault, you're in good company.

Anyway, the over-simplified version goes as thus: A UNC2814 co-opted Google Sheet file is used to connect to a Google Service Account for API authentication, before wiping itself and allowing its attackers backdoor access via a 16-byte cryptographic key "present in a separate file on the host at the time of execution."

"Once the Sheet is prepared, the backdoor conducts host-based reconnaissance. It fingerprints the endpoint by collecting the victim’s username, endpoint name, OS details, local IP address, and environmental data such as the current working directory, language settings, and local time zone." says Google.

"This information is then exfiltrated and stored in cell V1 of the attacker-controlled spreadsheet."

The access can then be used to transmit shell commands and mask the transfer of data to "identify, track, and monitor persons of interest."

At least, that's what Google believes it was doing: "We expect UNC2814 used this access to exfiltrate a variety of data on persons and their communications. Similar campaigns have been used to exfiltrate call data records, monitor SMS messages, and to even monitor targeted individuals through the telco’s lawful intercept capabilities."

"GTIG did not directly observe UNC2814 exfiltrate sensitive data during this campaign. However, historical PRC-nexus espionage intrusions against telecoms have resulted in the theft of call data records, unencrypted SMS messages, and the compromise and abuse of lawful intercept systems", says Google. Okey-dokey then.

Anyway, beyond what UNC2814 did or did not get away with, according to Google it's been thoroughly disrupted—and presumably sent to bed with no dinner. Google Sheets has now entered the witness protection program, and is believed to be on the mend. Oh okay, I'll finish off with something serious, if I must. Per the conclusion of the Google blog:

"The global scope of UNC2814’s activity, evidenced by confirmed or suspected operations in over 70 countries, underscores the serious threat facing telecommunications and government sectors, and the capacity for these intrusions to evade detection by defenders.

"Prolific intrusions of this scale are generally the result of years of focused effort and will not be easily re-established. We expect that UNC2814 will work hard to re-establish their global footprint." Dun-dun-duuuuuuun!

As the AI boom continues to, well, boom, it's not surprising that some are questioning the viability of rampant AI adoption, and the pace at which the hardware supply chain underneath it has to keep up.

One such commenter on the subject is none other than the product lead for Google AI Studio, Logan Kilpatrick. Posting on X, Kilpatrick said: "The compute bottleneck is massively under appreciated."

"I would guess the gap between supply and demand is growing single digit % every day", Kilpatrick continued. "In practice, this will be the rate limit on the impact AI will have in the economy/society."

Certainly, it's not the first time we've heard Google's name in association with massive, seemingly unsustainable AI compute demands. A leaked internal slide shown during a company meeting last year informed one of its teams that AI compute "must double every six months... the next 1000x in 4-5 years."

Simply put, there's only so much compute to go around, and there's a limit to how quickly it can be realistically expanded, even for Google.

TSMC, the manufacturer behind 90-95% of the world's most advanced chips, already looks to be near maximum capacity, resulting in Nvidia CEO Jensen Huang giving the company the hurry up in public.

More facilities look to be on their way—but the idea of it catching up with current demand any time soon seems unlikely. And let's not even mention the DRAM supply crisis, and what that's doing to electronics pricing right now.

So, without major efficiency gains on the software side of things, and with more and more AI tools, agents, and doohickies demanding vast amounts of computing capacity to operate, there seems to be an obvious bottleneck as to how fast the supply chain can keep up. Even if you do end up building data centers in tents.

X user @zazmic has many of the same thoughts: "Honestly, everyone talks about AI changing the world, but if even Google is running into compute limits, that tells you something... you can have all the ideas in the world, but if there aren’t enough chips to run them, progress slows down whether you like it or not.

"AI isn’t just a software story, it’s a supply chain story now, and that simply doesn’t scale overnight."

However, a built-in limiting factor to the mass adoption of compute-heavy AI agents might not be the worst thing.

A report released by Citrini Research earlier this week, which posited a doomsday-like scenario in which the widespread use of autonomous AI agents caused the US economy to falter (and eventually leads to mass protests on the streets), seems to have shaken US market confidence in the tech to a surprising degree.

Speaking personally, I can say that the idea of a built-in limit on how fast AI can expand is somewhat cheering. While the current demand is ruining the availability of PC gaming hardware (and certainly won't be helped by further supply/demand woes), perhaps a built-in limit means I'll be able to keep writing articles for a while yet, yes?

Small wins, folks. I have to take the small, intensely personal wins.

The AI-fuelled memory supply crisis has been difficult to escape—after all, even those literally living under a rock still need RAM for the glowing RGB rig they're keeping under there.

But it's important to note AI datacentres aren't gobbling up consumer-grade kit specifically. Some sources suggest major players have bought up close to 40% of wafer production (i.e., not fully-finished chips). Others are specifically buying up High Bandwidth Memory (HBM), which is specifically built for the demands of an AI datacentre—and Samsung may soon be raking it in thanks to its "industry-first commercial HBM4."

Samsung's asking price for this oh-so-fresh hardware? About $700. As Yonhap News points out, that's a 20–30% price increase over the last generation HBM3E (via Jukan).

Samsung can get away with charging this eye-watering price tag for a number of reasons. An industry insider told Yonhap News, "With commodity DRAM profitability now exceeding that of HBM, Samsung no longer has a reason to maximize HBM4 volumes at the expense of its commodity DRAM lines.

"Having proven its HBM competitiveness recovery through best-in-class performance and industry-first mass production shipments, Samsung is now reportedly adjusting capacity carefully from a profitability standpoint."

So, in other words, Samsung knows both DRAM in general and HBM4 are in high demand—but it's not going to make oodles more to meet that demand. This makes sense beyond just profitability; should the market turn, Samsung won't be left holding a whole load of expensive HBM it can no longer sell or production facilities it no longer needs.

Currently, industry sources suggest that only Nvidia has requested HBM4 so far this year. Other heavy hitters, like Google, are apparently still focused on securing HBM3E for its AI acceleration plans. With a price tag like that, who could blame them?

There's a strong possibility that Nvidia is currently preparing to unveil AI accelerators kitted out with Samsung's shiniest HBM during the GTC 2026 conference in March. Whether that includes Nvidia's Vera Rubin, the six-trillion transistor 'superchip' for AI, remains to be seen.

But while we're on the subject, rumours claim Rubin will be first in line to receive TSMC's ultra-advanced A16 chip node. As this year's GTC is slated to begin in San Jose on March 16, we won't have to wait too much longer to find out.

And as for us gamers, HBM isn't really our bag. AMD did try it with the RX Vega cards once upon a time, but that generation didn't go down well with the gaming masses. We'll stick to our very limited supply of GDDR, thanks.

AI has been in a funny place for some time. Growing exponentially, and sucking up much of the world's supply of memory, GPUs, and energy in the process, we've seen countless examples of models being released with few guardrails and being reigned in after the fact.

This week, Google published its "responsible AI progress report" (PDF warning), outlining why its "AI Principles are the north star standards" that will guide it over the next year.

"As models grow even more sophisticated, we see users and businesses around the globe transitioning from exploration to integration, finding new ways to put these tools to work in their daily lives" the company says.

Google notes the introduction of AI tools in 'scientific discovery', as well as 'clinical milestones in health' and even cites 'vibe coding' (which was Collins Dictionary's word of 2025).

This is all to build towards the central argument that AI isn't just being tested, but has thoroughly wormed its way into everyday society, and there's no going back. "The AI era is no longer a distant promise; it is a present reality that is beginning to unlock extraordinary opportunities for society."

The progress report is mostly centred on what AI is currently doing to aid humans, and how Google is attempting to make its Gemini chatbot safer. It reports that Gemini 3 saw "rigorous testing" prior to its deployment, and ways that Google is attempting to make it safer as it begins to "introduce agentic capabilities to Chrome."

Prompt injection, where bad actors feed nefarious instructions to LLMs to get models to exploit their own users, is a focus for future security advances from Google. This includes the addition of an "Alignment critic" that "acts as an independent reviewer, vetoing actions that do not align with the user’s specific intent."

Agentic AI is another focus for Google in 2026. Citing research done in April last year, Google argues, "The research assumes that highly capable AI could be developed by 2030." As AI becomes more powerful and gets further ingrained into PCs, it is more targeted by bad actors, so Google claims it is considering "various mitigations, such as blocking access to dangerous capabilities by using filters to prevent misuse, or using AI assistance to help maintain oversight."

Though Google is very focused in this report on mitigating harm, it appears to be just as focused on proselytising the benefits of AI. Google is reportedly attempting to:

• Accelerate scientific progress through research and an 'AI co-scientist tool'
• Improve global health through disease detection and the lightening of administrative burden
• Strengthen resilience with models designed to accurately read the weather and predict floods, cyclones, and earthquakes
• Support education through developing personalised learning plans and teaching AI literacy

Given the huge costs of AI and how much of the world's economy has been sunk into it, you'd imagine that any company committing to AI would really have to believe in it. Or at least say it believes in it. I suppose arguing for its use case in medicine and education is certainly stronger than 'because I can generate a Studio Ghibli-style photo of myself.'

NotebookLM, Google's notetaking tool, has an AI-generated voice feature that is surprisingly close to a real voice. One radio host argues it's not just close, but an actual ripoff of their voice, and has taken that claim to court.

Former NPR radio host David Greene filed his case in California on January 23, and his lawyer argues, "This case arises from Google’s deliberate acts of theft. Google used Mr. Greene’s voice without authorization and then used those stolen copies to develop, train, and refine its AI broadcasting product, NotebookLM."

According to The Washington Post, Green said, “I was, like, completely freaked out… It’s this eerie moment where you feel like you’re listening to yourself.”

Citing Greene's work with NPR shows, but also political radio show Left, Right & Center, the filing argues

"Google sought to replicate Mr. Greene’s distinctive voice—a voice made iconic over decades of decorated radio and public commentary—to create synthetic audio products that mimic his delivery, cadence, and persona."

David Green's voice.

It argues that podcast and radio hosts are typically paid for the use of their voices, as it adds credence and legitimacy to shows, which Google can effectively bypass with a similar voice.

This is something I noticed with the tool back in 2024. The podcast hosts confidently announce the facts, which is fair and probably good when they're actually facts. However, the fact that you can plug in any source you like means it can add that confidence to pretty much any idea or theory.

Greene's representative argues "Failure to pay the negotiated and agreed-upon price for such professional services, is a violation of multiple statutes and common law", and that "Defendant Google is attempting to disrupt the podcast industry."

The case itself compares a clip of David Greene talking about Trump's Big Beautiful Bill to an AI-generated summary of Greene's analysis (alongside a few others). You can listen to those clips below to see for yourself how similar they might be.

An excerpt from David Greene's coverage of the Big Beautiful Bill

NotebookLM's analysis of David Greene's clip

Notably, there doesn't appear to be firm proof that Greene's voice was used in Google's training data. Greene's team notes "an independent forensic software company specializing in voice recognition" analysed the two voices, and "the tests indicated a confidence rating of 53-60% (on a -100% to 100% scale) that Mr. Greene’s voice was used to train the software driving NotebookLM."

In conversation with The Washington Post, Adam Eisgrau, the Chamber of Progress's Senior Director of AI, Creativity and Copyright Policy, says, “If a California jury finds that the voice of NotebookLM is fully Mr. Greene’s, he may win. If they find that it’s got attributes he also possesses, but is fundamentally an archetypal anchorperson’s tone and delivery it learned from a large dataset, he may not.”

Google has responded to this suit to the Washington Post, calling the allegations "baseless".

José Castañeda, a Google spokesperson, said “the sound of the male voice in NotebookLM’s Audio Overviews is based on a paid professional actor Google hired.”

This mimics a similar case from 2024, where Scarlett Johansson noticed GPT-4o's "her" voice was "eerily similar" to her own. OpenAI soon paused the use of the voice, but it looked particularly strange, as Johansson argues OpenAI previously approached her to add her likeness, which she declined.

As Green's case was only recently filed, and we don't have any proof that Greene's voice is in the data set, we don't yet have an indication of how it will go. Either a win or a loss will play into a larger precedent over which data AI companies can and can't use, and how different the output needs to be as a response.

As AI continues to grow and make its way into everyday life, the alleged productivity gains do appear to be showing in some places. It just so happens that hacker groups are one of those places, and Google's Threat Intelligence has listed some of the many ways they use it. Welcome to the future.

In its latest report, it says, "In the final quarter of 2025, Google Threat Intelligence Group (GTIG) observed threat actors increasingly integrating artificial intelligence (AI) to accelerate the attack lifecycle, achieving productivity gains in reconnaissance, social engineering, and malware development."

It details that government-backed threat actors, like those reportedly in the Democratic People's Republic of Korea (DPRK), Iran, the People's Republic of China (PRC), and Russia, are using LLMS for "technical research, targeting, and the rapid generation of nuanced phishing lures".

One of the bigger growing threats is called a model extraction attack. In Google's case, this involved accessing an LLM legitimately, then attempting to extract information to build new models.

Google reports one case of over 100,000 prompts which were intended to emulate Google Gemini's reasoning capabilities. Naturally, this is more of a threat to companies than to the average user. However, there are more methods detailed in the report.

One such method for AI use is making hackers seem more reputable in conversation. "Increasingly, threat actors now leverage LLMs to generate hyper-personalized, culturally nuanced lures that can mirror the professional tone of a target organization or local language"

Google has spotted it being used in phishing scams to learn information about potential targets, too. "This activity underscores a shift toward AI-augmented phishing enablement, where the speed and accuracy of LLMs can bypass the manual labor traditionally required for victim profiling."

This is all before mentioning AI-generated code, with hackers such as APT31 using Gemini to automate analysing vulnerabilities and plans to test said vulnerabilities. It also spotted 'COINBAIT', a phishing kit masquerading as a cryptocurrency, "whose construction was likely accelerated by AI code generation tools."

Though mostly a proof of concept, Google has reportedly spotted a malware that prompts users' AI bots to create code to generate additional malware. This would make tracking down malware on a machine increasingly hard as it continues to 'mutate'.

Google says, "The potential of AI, especially generative AI, is immense. As innovation moves forward, the industry needs security standards for building and deploying AI responsibly."

Just last week, we saw a phishing scam that uses AI to deepfake CEOs of companies, in order to get access to a victim's cryptocurrency. It seems AI is becoming more than just one tool in a hacker's toolbelt, and one has to hope counteragents are getting enough data to counteract it.

A hacking group reportedly based out of North Korea has come up with a "new tooling and AI-enabled social engineering" scam, according to Google, and it's pretty complicated.

Effectively, it uses a hacked account to send a Zoom link via a calendar invite to an uncompromised account. That version of Zoom is, in fact, a spoof, and what targets are met with is a deepfaked version of the account owner. Google's report notes that a version of this deepfake takes the form "of a CEO from another cryptocurrency company."

Once in the meeting, the deepfaked user claims to have technical issues and directs the target on how to troubleshoot their PC. The troubleshooting prompt leads them to run an infected string of commands that then unleashes a series of backdoors and data miners on the victim's PC.

Google calls it "AI-enabled social engineering" and notes 7 new malware families used in the attack.

UNC1069 are the actors Google has identified as being behind the scam. They have reportedly been active since 2018 and were found to have been using Gemini last year to "develop code to steal cryptocurrency, as well as to craft fraudulent instructions impersonating a software update to extract user credentials".

Google says UNC1069 is "employing these techniques to target both corporate entities and individuals within the cryptocurrency industry, including software firms and their developers, as well as venture capital firms and their employees or executives."

This hack needs access to an account to start in the first place, so Google notes further attacks have "a dual purpose; enabling cryptocurrency theft and fueling future social engineering campaigns by leveraging victim’s identity and data."

Though Google states that the account linked to the group has been terminated, Gemini was used at some point "to develop tooling, conduct operational research, and assist during the reconnaissance stages."

Gemini is not the only AI tool being used in similar cybercrimes. Antivirus creator and cybersecurity company Kaspersky claims hacking group BlueNoroff is using GPT-4o to enhance images to convince targets.

As AI gets more impressive and complicated, so too will the scams to accompany it. One can only hope that anti-scam measures become equally clever.

Google and Alphabet CEO Sundar Pichai has handily summarised the latter's Q4 results in a blog post this week, and right near the top is an interesting figure: YouTube's annual revenues surpassed $60 billion across ads and subscriptions.

Looking deeper into the fiscal results themselves, YouTube ads brought in $11.38 billion in the last quarter, compared to $10.47 billion in last year's report. Anecdotally, I can well believe this figure, as YouTube's ad frequency ramped up at such a rate towards the end of 2025 that I finally caved and bought a YouTube Premium subscription myself.

Speaking of which, while Alphabet doesn't report YouTube Premium's revenue individually, it has declared that the company now has "over 325 million paid subscriptions across consumer services, led by strong adoption for Google One and YouTube Premium."

On the surface, YouTube's revenue figure seems impressive—but its viewing figures are even more so. According to CEO Neal Mohan in his 2026 YouTube community letter, YouTube Shorts alone average out to 200 billion daily views. The YouTube head honcho also reports that YouTube has been "#1 in streaming watchtime in the U.S. for nearly three years", according to Nielsen's The Gauge report.

It's difficult to get reliable data on YouTube's usage figures overall, but it's said to have between 2.5 and 2.7 billion monthly active users worldwide as of 2025, with well over 1 billion hours of video content consumed daily, and over 20 million videos uploaded on average per day. If those figures are even close to accurate, that's an astonishing amount of data to handle—even for a cloud behemoth like Google with its own vast server infrastructure to lean on.

Likewise, it's very difficult to get a bead on exactly how much YouTube costs to run. But once creator monetisation payments, employee costs, and server maintenance are brought into the equation, it must be a pretty penny. Alphabet's Q4 financial report says that the revenue of Google Services in total was $95.86 billion, with its operating income reported as $40.132 billion, so that's $55.73 billion spent on servicing costs in the last quarter alone.

Of course, that's including Google's vast services infrastructure overall, not just YouTube. Still, it does make me wonder whether YouTube in and of itself is as profitable as its parent company hopes it would be, despite the impressive revenue figure. It's certainly more than Netflix, though, which reported $45.18 billion of total annual revenue earlier this year—along with operating expenses of $31.85 billion.

The ad strategy I (and many others) have observed in recent months certainly seems to suggest that YouTube is doubling down on its income potential, whether that's advertising or pushing people towards its paid premium services, or a combination of the two. It's remarkably good at noticing adblockers these days, too, which feels like an escalation of its previous, adblock-hating ways.

It's all about money at the end of the day. Still, with so many eyeballs on its videos, and so many excellent (and many dubious) creators using its platform, I will admit that YouTube Premium has finally got me to open my moth-ridden wallet and cough up. What can I say, I like a bit of ad-free YouTube before bedtime. Don't we all?

Google has warned that well-known and already-patched exploit for the WinRAR file archiving and compression tool for Windows remains in "widespread, active" use by "government-backed threat actors linked to Russia and China".

Known as critical vulnerability CVE-2025-8088, the exploit identified was in July last year and was posted on the National Vulnerability Database back in August. It's widely known and numerous other bodies, even including the UK's NHS, have registered the threat.

The exploit was actually addressed by the makers of WinRAR, RARLAB, with the 7.13 update on July 30 last year. Of course, that isn't going to help anyone running earlier versions of WinRAR.

As we understand it, the exploit works by concealing a malicious file within within the ADS of a decoy file in a WinRAR archive. When a user extracts the archive, the payload is saved to critical locations such as the Windows Startup folder via path traversal sequences and then automatically executes upon a machine restart.

Google says the bad guys involved include such favourites as "Russia-nexus" actors targeting the Ukrainian military, China-nexus actors exploiting the vulnerability to deliver the POISONIVY malware via a BAT file dropped into the Startup folder, which then downloads a dropper, and financially motivated hacking groups.

Among the latter, Google says one group targets hospitality and travel sectors using phishing emails around hotel bookings. Google concludes that this WinRAR bug just goes to show the "enduring danger posed by n-day vulnerabilities."

N-day vulnerabilities, of course, are known security flaws for which patches or fixes exist. The point being, again, that patches are only of any use with actual, ya-know, use.

All of which means the conclusion here is fairly straightforward. Happily, it's very easy to ensure you aren't at risk from this exploit.

If you use WinRAR and haven't updated to the latest 7.13 build, do that immediately. Until then, do not pass go. Do not open any WinRAR archive, no matter its provenance. And that's really it.

Mozilla has launched a new 'Pioneers' program that's looking to pay people to "work with our New Products leaders to build tools and products for the next version of the web" for a short amount of time. The idea seems to be to lean into a shift in direction that the company's been talking about for quite some time, towards a more open but more AI-centric web.

Well, actually, I'll let Mozilla's New Products SVP, Peter Rojas, tell you: "Our mission at Mozilla is to ensure the internet is a global public resource, open and accessible to all." Selected Pioneers will work on new products that help move towards this, as the program is for "a supporter of Mozilla and its mission as the end goal is to build with us full time."

In addition to gelling with its controversial decision to have Firefox "evolve into a modern AI browser", this also aligns with the company's recent announcement of its "open-source AI strategy." Closed AI systems, Mozilla's CTO Raffi Krikorian says, are currently winning: "I understand the appeal firsthand, because I’ve made the same choice myself on late-night side projects when I just wanted the fastest path from an idea in my head to something I could actually play with.

"What we’re dealing with isn’t a values problem where developers are choosing convenience over principle. It’s a developer experience problem. And developer experience problems can be solved."

The solution, according to Mozilla, is to tackle the problem by improving developer experience so they can "build the open ecosystem themselves", giving people and communities who create data "a say in how it's used and a share in the value it creates", moving towards edge rather than centralised AI models, and trying to tackle the compute "choke point".

It's the former two that the company is working on right away, though, and to this end, in addition to some more general and vague directives, the company currently has a couple of concrete programs.

The first, any-suite is a "unified open-source stack that simplifies building and testing modern AI agents and apps", which fits right into what Mozilla seems to be going for, vying for its own open-source stack to compete with the current dominant proprietary ones. Presumably, some of the Pioneers that Mozilla is looking for will be ones that work towards tools that slot into the any-suite or at least complement it.

The second relevant program is the Mozilla Data Collective, which is a collection of AI training datasets "built by and for the community in a transparent and ethical way … A marketplace for data that is properly licensed, clearly sourced, and aligned with the values of the communities it comes from."

This all rings a lot better to my ears than "AI browser" does, although I suppose this is all apart from Firefox, concerning new products. That might mean less of a problem of funding, which Mozilla has long faced with Firefox.

Almost all of Firefox's funding comes from its partnership with Google, which raises questions about just how free Mozilla is to make meaningful changes to the browser. That might not be the case here, though, with new products and a whole AI stack.

I suppose there's an element of techno-deterministic 'if it's happening, it might as well happen ethically' reasoning slipping through the cracks in my thinking here, too, though. Who says the next version of the web has to be AI-centric?

Still, it is a positive step to move towards open-source software and more ethical data scraping and use. If it follows through, Mozilla will certainly be in good company. The EU has recently started pushing towards open-sourcing Europe's software, and I'm sure Wikipedia will be happy with Mozilla's focus on transparent and ethical datasets, given its recent focus on pushing for proper data and content attribution.

I enjoy a good traced ray, AI-upscaled texture, or 'hallucinated' frame as much as the next person, but I think pixel art remains an important part of any art team's tool kit. Pixel art offers far more than a nostalgic throwback appeal, capable of a level of expression that would give even the most photorealistic polygons a run for their money. So when I stumbled across the Isometric NYC city map project, I was at first properly stoked—and then deeply conflicted.

Isometric NYC is the one-man effort of Andy Coenen, attempting to render all of New York City in an isometric pixel art style distinctly reminiscent of the SimCity series. But rather than laboriously building this city pixel-by-pixel, Coenen has instead leveraged multiple AI agents to create this city map. The entire process required much more work than dashing off a handful of prompts, but you are perhaps beginning to understand my sense of conflict.

Coenen himself is a software engineer currently conducting AI research at Google DeepMind. He explains that the original idea for Isometric NYC was to use Google's "Nano Banana to generate a pixel art map from satellite imagery tile-by-tile," though Coenen ultimately used a number of AI agents, including "Claude Code, Gemini CLI, and Cursor." He also shares that he "wound up writing almost no code for this project."

He first set to work by 'whiteboxing' city geometry using 3D CityGML data. Unfortunately, his original plan didn't work out because "there was a bit too much inconsistency between the 'whitebox' geometry and the top-down satellite imagery, and Nano Banana was prone to too much hallucination in resolving these differences."

Moving to Google Maps 3D tiles API's more precisely aligned geometry and texturing, Coenen encountered further issues with Nano Banana. Being such a large generative AI model, Nano Banana struggled to output a stylistically consistent city, and ultimately proved both too slow and too pricey for the project's intended scope. So, Coenen elected to train something smaller and cheaper.

By feeding a Qwen/Image-Edit model "a training dataset of ~40 input/output pairs," Coenen began to get AI-generated city tiles closer to his preferred style in a fine-tuning process that took "~4 hours and cost ~12 bucks." Many more weeks of work followed to create software that would allow you to zoom in and out of each tile without completely hammering your hardware.

At a glance, the final result is pretty breathtaking—even that billionaire apartment building everyone hates, 432 Park Avenue, is represented here. The illusion falters when you zoom all the way in; there's a telltale, AI goopiness that immediately dispels any illusion of purposefully placed pixels.

Now, on the one hand, creating all of this by hand would have taken a lifetime of work for one person, and at that point, what would have to sustain you throughout such a project is pure love of the pixel art game. That's all without getting into whether you decide to stay on top of representing New York City's constant stream of changes, or simply choose to let every pixel art building be its own time capsule. Coenen's method has produced a somewhat accurate chunk of New York in an isometric-inspired style on a comparatively shorter timeline.

It's important to note, though, that Coenen isn't motivated by the business-brained efficiency mindset we've perhaps come to expect from creative applications of generative AI. He shares, "I spent a decade as an electronic musician, spending literally thousands of hours dragging little boxes around on a screen. [...] This isn't creative. It's just a slog."

He acknowledges that this 'slog' can also hone your creative instincts, but still feels AI could represent 'the end of drudgery' in creative fields. Though interestingly, he ends that train of thought thus: "If you can push a button and get content, then that content is a commodity. Its value is next to zero. Counterintuitively, that’s my biggest reason to be optimistic about AI and creativity. When hard parts become easy, the differentiator becomes love."

And to me, that sounds like the Google DeepMind AI researcher agreeing that love of the game is key to art and human creativity.

Former PC Gamer writers have gone on to do amazing things: Pen Star Wars movies, sell millions of extremely good videogames, and, maybe, save the internet? I exaggerate slightly, but tech writer Corbin Davenport released a tool last week called Just the Browser that offers the faintest glimmer of light in this AI abyss we now call the web.

Per the description on Just the Browser's Github page, the project is a script that, when run, will "remove AI features, telemetry data reporting, sponsored content, product integrations, and other annoyances from web browsers." On Windows, all you have to do is open PowerShell as administrator and copy-paste this command:

& ([scriptblock]::Create((irm "https://raw.githubusercontent.com/corbindavenport/just-the-browser/main/main.ps1")))

The script will then ask you which browser you wish to de-AI, with support for Chrome, Firefox and Edge. Press a couple keys and you're done. It's just as simple to run in the terminal on Mac or Linux.

To set expectations, Just the Browser does not touch anything on the actual websites you visit, which means you're still going to be seeing AI crap all over the place. It is not a silver bullet. What it's doing is using group policy settings—the sorts of under-the-hood levers that IT folks configure that aren't typically visible to the everyday user—to disable features that don't have an easy off switch. It's not a hack or a browser extension, and you could do the same thing manually if you wanted to—Corbin just bundled up a bunch of nice little changes into one simple script.

If you're a Chrome user, here's all the stuff Just the Browser is disabling, which you can see for yourself if you type chrome://policy/ into your address bar.

The open source project has gotten a blitz of attention this week, popping up on Hacker News and a range of tech sites. This has of course invited a number of correct-but-nonetheless-obnoxious commenters pointing out that running a random PowerShell script you find on the internet is very bad opsec, and who says we should trust this random guy anyway? I am! I'm saying it!

Corbin has a long history of making cool, free open source projects, and Just the Browser also has a documented manual process if you want to see exactly what the script is changing (which is just the stuff in the table above).

Great work, Corbin. Now we just have to figure out how to block the rest of the slop.

Seems like only the other day I was chatting about yet another big tech monopoly—because I was. But instead of Google appealing the US Department of Justice's antitrust ruling against it, this time the Federal Trade Commission is appealing a November 2025 ruling concerning Meta.

The FTC originally filed its lawsuit against the company, then called Facebook, all the way back in 2020. The original suit alleged the company had "engaged in a systematic strategy" of buying up competitors, such as Instagram and WhatsApp, in order to maintain its social media monopoly.

The case was brought before a US district judge in Washington, who ultimately ruled last year that Meta had not violated antitrust laws with these acquisitions. In his decision, Judge James Boasberg even noted that the FTC itself had reviewed and approved both Meta's 2012 acquisition of Instagram, as well as the company's 2014 purchase of WhatsApp.

Undeterred, the FTC is appealing that decision, and still highlights those same acquisitions as the motivation behind its most recent appeal.

FTC Bureau of Competition Director Daniel Guarnera issued the following statement: "Meta has maintained its dominant position and record profits for well over a decade not through legitimate competition, but by buying its most significant competitive threats. The Trump-Vance FTC will continue fighting its historic case against Meta to ensure that competition can thrive across the country to the benefit of all Americans and U.S. businesses."

This is also far from the only legal hot water Meta finds itself in. Just in terms of antitrust action, Italian authorities have also launched an investigation into Meta. Beyond that, another court case last year alleged Meta violated copyright law by torrenting terabytes of data from 'shadow libraries' in order to train its AI products. Eventually, 13 book authors spearheaded this particular legal challenge, though Meta ultimately won the case before it could go before a jury.

As for how the aforementioned, acquired social media apps are faring under Meta, Instagram has ditched fact-checkers and seen a tidal wave of gore accidentally recommended to users, while 3.5 billion WhatsApp users saw their data scraped in what could potentially be the 'largest data leak in history.'

One of the most obnoxious traits of the AI slop era is the way any phrase you type into Google is now likely to prompt an AI overview speculating about what a sequence of words means, making up a bunch of stuff whole cloth, and being completely wrong.

Last week while trying to find traces of an internet forum I remembered from my teens, I typed in the site's slogan—"Grimmest and Most Frostbitten Kvlt of Fool's Gold"—which Google's AI overview spun up a completely nonsense answer for: "The grimmest, most frostbitten "kvlt" (cult) of fool's gold is Iron Pyrite (FeS₂)." Google further explained that the "'Kvlt aspect" was due to fool's gold being "often considered more valuable than real gold in the fictional universe of Azeroth (World of Warcraft)." Whatever, man.

The other most obnoxious trait of Google's AI search results is that when it does actually recognize what you're referencing, it will source a bunch of random information, assemble it into what looks like a factual overview, and still be completely wrong. Case in point: Author Chuck Wendig does not have a pet cat, but that didn't stop Google's AI overview from citing a blog in which he jokes about a cat named Sir Mewlington Von Pissbreath.

In a December post titled Vital Cat Update, Wendig expressed surprise that a Google search for "Does Chuck Wendig have a cat" was prompting an AI overview claiming he did have one such pet, named Boomba, according to the website "Wengie Wiki." (Astute readers will note that "Boomba" is not the same name as Sir Mewlington Von Pissbreath; this is because Google's AI overviews are even stupider and easier to manipulate than you might already think).

"You might be saying, 'Chuck, I didn’t know you had a cat!' and I’d respond with, 'I didn’t know I had a cat either,'" wrote the author of Star Wars novels and his own fantasy series'. Wendig bowed to the wisdom of the AI, which he called "totally not a piece of shit that just makes up information willy-fucking-nilly," and embedded several more search results in which the AI overview explained his cat Boomba had died, and that he had adopted a new cat named Franken. Google cited Wendig's blog for this information—curious, since a cursory search of the blog reveals he'd never used the words "Boomba" or "Franken" once.

The blog goes on to show that running multiple searches related to the names of his imaginary pets resulted in Google's AI overview continually making up new names and pets one after another. It also said that he had cancer, and had "embraced Christianity in a public way."

"This is just a nice little reminder that generative AI is shit," Wendig wrote. "Total shit! It scrapes everything we’ve ever written and then can’t even sort through it fast enough to give us a correct answer, all the while burning down the world to lie to us. What a truly nightmarish thing we’ve created! Jesus Christ we are cooked!"

Wendig ended the post by referencing his cat "who is named Sir Mewlington Von Pissbreath and who is definitely real and who is six years old and who wears a little top hat and also can speak limited Cantonese."

Six weeks later, Google's AI results have incorporated Sir Mewlington into the vast corpus of "facts" it will serve up, incorrectly, as it endeavors to make every search result page untrustworthy.

Good blog, though. Wendig further excoriated AI—specifically about its use in the publishing world—in another post in late December, which is soothing if you appreciate published authors saying things like "I think AI is only inevitable when we believe the lie of its inevitability." Maybe reading it will make you feel a tiny bit more sane in an insane world.

The saga of Google versus the US Department of Justice continues. On Friday, the multi-billion dollar corporation asked the Court to delay some of its proposed antitrust remedies while Google appeals the ruling.

To recap, back in 2024 U.S. district judge Amit Mehta ruled that Google had violated antitrust laws when it used exclusivity agreements to maintain its monopoly over online search. A number of remedies were proposed, including Google's divestment of the Chrome browser (which will ultimately not come to pass), and most pertinently right now that the search giant shares its data with rival companies.

The DOJ secured this latter remedy among others back in September, with the US District Court ordering "Google to make certain search index and user-interaction data available to rivals and potential rivals." Google is now asking to postpone the sharing of this data while it appeals the DOJ's ruling that it holds an illegal monopoly.

To date, many have criticised how the DOJ is handling its case against Google, saying that many of its remedies so far have amounted to little more than 'a slap on the wrist.' But Google insists data sharing would be a step too far at this juncture.

In court documents filed on Friday, the company also makes the case that sharing this data before hearing its appeal would "would irreparably harm Google because competitors will gain access to vast amounts of Google’s proprietary information." The company argues that, regardless of the result of its appeal, making this disclosure beforehand would reveal trade secrets it would have no way to recover (via Reuters).

Vice president of regulatory affairs at Google, Lee-Anne Mulholland, wrote in a blog post, that the sharing of Google's data would additionally "risk Americans’ privacy and discourage competitors from building their own products—ultimately stifling the innovation that keeps the US at the forefront of global technology."

Mulholland also wrote on the ruling more broadly, saying, "As we have long said, the Court’s August 2024 ruling ignored the reality that people use Google because they want to, not because they’re forced to."

To me, this feels especially disingenuous. To speak only about myself, I use Google products like Search because many of my corporate accounts already use GSuite. It's also how many folks read my work, so I have to at least have a professional understanding of Google search (you do not want to get me started on Google allowing its AI to take a hatchet to PC Gamer's carefully crafted Discover headlines).

But even outside of work, I chose an Android phone and its integration with the Google Play app store because I personally dislike Apple's pricey walled garden even more. It's arguably not much of a choice when it's at best a two-horse race.