Anthropic Resignation Ignites Washington's AI Safety Fight

Hubinger's public endorsement — the >10% extinction estimate — is what Fortune identified as the difference between this resignation and the four insider warnings before it, from Geoffrey Hinton in 2023 through Sharma in February.

Sen. Bernie Sanders and Rep. Greg Casar are proposing a permanent ban on "superintelligent AI systems," defined as systems that "match or exceed human cognitive performance and capabilities across a broad range of domains or tasks," plus a pause on advanced AI development until a new federal agency exists to regulate it.

Bores, who was an engineer at Palantir before politics, told Fortune that "concerns about AI have been widespread for a while, but a few industry players have been willing to spend hundreds of millions to silence elected officials."

On the right, Gateway Pundit columnist Drieu Godefridi argues the viral moment was staged: he writes that The Wall Street Journal published its exclusive 18 minutes before Coxon hit publish — "That is not a leak. That is an embargo." Godefridi claims the first amplifiers were a funded AI-safety network bankrolled by the same donors tied to Anthropic, that 22 Democrats piled on in a single day, and that Sanders had his bill written five days earlier. His conclusion: "Coxon didn't create the legislation. He just put a face on it." The Frank has not independently verified those claims, and Godefridi offers no evidence that Coxon coordinated with anyone.

Godefridi does not dispute the fear itself. "The fear inside the labs is real," he writes. "The scam is the machinery that turned a four-month employee into a 150-million-view political weapon overnight." His argument is that licensing rules would decide "who gets the permit" and kill open models — "Not Skynet. 1984."