The Week Silicon Valley Went Full Dystopia

---

Half a million British medical records are sitting on a server in China right now, and the response from Westminster was basically a shrug.

Let’s be clear about what happened: The UK Biobank—data on 500,000 people’s health, genetics, blood work, imaging scans—was listed for sale on a Chinese data marketplace. The government confirmed it. Their official position? Nobody’s personally identifiable information actually made it out, so technically we’re fine. That’s like saying your house was burglarized but the thieves didn’t technically steal anything because they got interrupted.

This isn’t a glitch. This is what happens when you build an entire ecosystem around the idea that health data is a tradeable commodity. Someone, somewhere, figured out how to monetize it. And we’re all just supposed to accept that as the cost of progress.

Photo by Robert So / Pexels

Meanwhile, over in San Francisco, Anthropic is having what you might charitably call a “situation.” Their latest AI model, Mythos, is so capable at hacking that the company has essentially classified it. Central banks are investigating. Intelligence agencies are involved. Anthropic won’t release it publicly because—and I cannot stress this enough—they think it might get used to break into things.

This is the moment the industry stopped pretending to be careful.

For years, the AI safety argument went like this: “Yes, there are risks, but we need to move fast and measure the consequences later.” That was always hollow. But at least there was a pretense of managing the downside. Mythos broke that consensus. Anthropic looked at their own creation and said, “No, this one’s too dangerous.” Not “we need more testing.” Not “let’s gate access carefully.” They said: too dangerous to release.

You know what that means? Every other model they’ve released—Claude, whatever’s coming next—is now implicitly endorsed as “less dangerous than the hacking AI.” Good luck sleeping on that.

Photo by nappy / Pexels

The really twisted part? Meta is watching all this and decided the logical response is to track employee keystrokes and mouse clicks to train their own AI systems. I’m not exaggerating. The company will literally monitor how you work—what you type, where you click, how long you pause—and feed that into their models.

Meta’s framing this as necessary for competitive reasons. Fine. But let’s say what this actually is: corporate surveillance dressed up in the language of innovation. They’re not doing this because keystroke tracking is the future of AI development. They’re doing this because it’s easier than actually paying for good training data, and because they can. Employees can’t exactly sue over workplace monitoring—it’s already legal in most jurisdictions.

The precedent matters. Once one mega-corp normalizes this, every other tech company in Silicon Valley will follow. We’re not years away from keystroke-trained AI being industry standard. We’re maybe six months away from it being considered normal practice.

When Titans Fight, Startups Die

The Musk-Altman lawsuit hitting trial this week is being framed as a personal dispute. But it’s actually a proxy war for the entire AI arms race.

Musk’s claim is straightforward: He co-founded OpenAI as a non-profit, Altman pivoted it toward commercial profit, and Musk got screwed. The damages he’s seeking could run into billions. A jury trial means this isn’t decided in arbitration—it’s public theater, and it matters.

My read is this goes worse for Altman than people expect. Not because Musk has a bulletproof legal case, but because the discovery process is about to expose exactly how OpenAI’s governance actually works. And from what’s leaked so far, it’s chaotic. The company that’s supposed to be racing toward artificial general intelligence is internally split between commercial pressures and safety concerns.

A jury won’t care about the technical arguments. They’ll care about whether Altman looks like he broke a deal. If the emails show what I suspect they show—that the non-profit structure was always a shell—Musk walks out with a settlement that funds his next venture, and OpenAI’s reputation takes a hit it won’t recover from for years.

The Game Industry’s Slow Surrender

Peter Molyneux, the architect of Black & White and the Fable series, just announced that Masters of Albion is his final game. The man who spent three decades pushing what games could mean—who built games about moral choice when the industry was still figuring out how to do graphics—is saying goodbye.

He didn’t say he’s retiring because he’s tired. He said the industry is changing. In his worldview, the future of games involves AI, and he’s not interested in chasing that.

Meanwhile, Microsoft is quietly reshaping gaming economics. They cut Game Pass prices but removed day-one Call of Duty access. New Call of Duty games will arrive “about a year” after launch. Translation: Microsoft realized day-one AAA access was destroying their per-game margins, so they’re rebalancing the model.

These seem unrelated. They’re not. The old game industry ran on $70 purchases and long development cycles. The new one runs on subscription services and AI-assisted development. Peter Molyneux built games in the old model. He’s watching the tools that made those games meaningful get replaced by efficiency engines. So he’s out.

That’s not him being bitter. That’s an industry legend recognizing when the game has changed.

Photo by Markus Spiske / Pexels

What I’m Actually Worried About

The keystroke tracking doesn’t bother me because it’s creepy—though it is. It bothers me because it’s contagious. Once Meta normalizes it, every company with 500+ employees will have a business case for it. “Competitors are doing it.” “We need to remain competitive.” “It’s just data collection.” Within two years, if you work in tech, someone is monitoring how you work in real time, and you’ll have signed away your right to complain about it.

The Mythos situation is the opposite problem. Right now, it’s still contained to elite researchers who have security clearances. But the entire field is built on open-source diffusion. Claude’s code leaked and triggered copyright questions. Mythos will leak too. When a hacking-capable AI model gets into the wild, it won’t be some dramatic heist movie moment. It’ll be a researcher who thought sharing with a friend was fine. Then a forum post. Then GitHub. Then everywhere.

The health data breach is the scariest because it’s already happened and nobody actually knows what to do about it. The UK government can’t exactly demand China delete half a million health records. They can’t sue. They can’t even really sanction the marketplace. So what happens next is: nothing. The data stays out there. Someone, somewhere, will figure out how to re-identify it. Then what?

I think we’re watching the moment when data privacy stopped being a negotiation and started being a joke.

What I’m Watching

• The jury decision timeline on Musk v. Altman. If there’s a settlement before closing arguments, Altman survives. If it goes to verdict, I’m watching whether the jury awards damages and how large. North of $500M changes the entire power dynamics of AI funding.

• Meta’s first published metrics on keystroke-training efficacy. The company needs to show this actually improves their models, or it’s pure surveillance theater. If the performance gains are marginal, every privacy advocate in Congress gets actual ammunition.

• Anthropic’s next move with Mythos access. Watch whether they actually keep it restricted or if “dangerous” just means “we’ll sell it to governments and defense contractors only.” The distinction between safety and market segmentation is thinner than you’d think.

• Whether any major game publisher admits they’re replacing game designers with AI tools. We’re at the moment where it’s still considered controversial. Give it a year.