Are the Doomers Now Doomed? A Bad Week for the AI Apocalypse.

Last week we were panicking about a supposedly imminent AI apocalypse. Seven days later, Keith Teare argues, government oversight is about as far away as it could possibly be. That’s because Musk, Altman and Amodei agreed to police themselves and test each other’s models. The doomers, says the publisher of That Was The Week, have lost not just this week’s battle but probably the whole war.

 

So who will then write the AI rules? The people building the models, Keith says — as with California’s building codes, government sanctions the standards but engineers write them. And trust doesn’t come from authority, it comes from use. The real check on our new AI overlords, he promises, will come in court, with the American legal system holding OpenAI and Anthropic financially accountable for all damage. That should doom the doomers. If we survive the apocalypse, of course.

 

Five Takeaways

 

•       The Week Regulation Went Away. Seven days, and the conversation reversed. Last week: the whistleblower, the headlines about AI killing everyone within a decade, the calls for global regulation. This week: Trump calling Jensen Huang live from an AI conference to make policy on the fly, Vance at All-In, and — as early as Monday — Musk, Altman and Amodei all agreeing that they are responsible for their own development, that they will slow model-building to match their control systems, and that they will begin testing each other’s models. Keith’s reading: the call for government oversight is going to fade. The New York Times map of the fight — accelerationists (Huang, Sacks, Andreessen), regulators, and a middle of Hassabis, Nadella, Pichai and Zuckerberg — holds at the edges but not the centre, which he calls shifting sands: Hassabis and Amodei made a proposal to the White House that was rejected this week; Zuckerberg then came out against government stepping in. And you cannot disambiguate self-interest from philosophy in that group: they have hearts, but they are paid to represent shareholders, and right now the rise of open source suits Google, Microsoft and Facebook.

•       Who Writes the Rules. Cohere’s Aidan Gomez asked whether a handful of market-dominant Silicon Valley companies should get to define the safety standards of a generational technology. Keith’s answer is an unembarrassed yes — and his analogy is the building code: government sanctions the rules, but the tolerances come from the architects and civil engineers who do the building, not from a committee of legislators. Which is why he thinks the Washington hearings shouldn’t be happening at all: time-wasting, pulling important people away from their work to talk to politicians whose primary interest is reelection. Government, he says, has no role yet — it is too early in the life of AI for anyone in Congress to understand it well enough to legislate. Andrew pressed the obvious objection: so we simply trust Sam, Dario and Elon to police themselves?

•       It Is You, Not the AI. The hour’s best answer, and the one that moves the argument on. It isn’t about trust, Keith says — it’s about accountability, and the machinery already exists. It would not be okay for their software to break into a bank and steal money, or into a rival and steal intellectual property; the legal system covers that today. “Up until now they’ve been saying, hey, it’s not us, it’s the AI. Well, actually, no. It is you.” And the nice thing about liability is that it makes recklessness expensive. Palantir’s CEO pushed the same logic to its conclusion on CNBC: the liabilities facing Anthropic and OpenAI are so enormous that only a government could carry them, so nationalize. Keith’s own caveat is the honest one: when he uses agents he is happy with the outcomes but cannot see the steps, and