Hassabis proposes an AI regulator as states move against data centers

Google DeepMind's Demis Hassabis calls for a U.S. standards body to test frontier models before release, a leading AI researcher brings his warnings about catastrophic risk to a mass YouTube audience, and New York becomes the first state to pause data center permits.

Demis Hassabis, CEO of Google DeepMind, proposes a majority independent U.S. standards body modeled on the Wall Street regulator FINRA, which would set benchmarks, decide which models count as frontier class, and test new models 30 days before release for cyber, bio, and deception risks. (Robert)
The dispatch argues that Hassabis undercuts his own plan: he describes a race outpacing our understanding, mentions coordinating a slowdown only in vague passing, and offers no answer for how one country's rules would apply to models built elsewhere. (Robert)
AI researcher Daniel Kokotajlo spent two hours on the YouTube show The Diary of a CEO discussing his team's AI 2040 scenario, his 70 percent estimate of a catastrophic outcome, and the equity he gave up rather than sign OpenAI's anti disparagement clause. (Robert)
Kokotajlo urged listeners not to dismiss AI dangers for sounding like science fiction, and said it is not too late to change course. (Robert)
New York became the first state to pause data center permits, a one year executive order covering facilities that use at least 50 megawatts, while other governments diverge and a March Gallup poll found 71 percent of Americans oppose data centers in their own community. (Donald)