Anthropic insider’s viral resignation reaches congress

A resignation letter posted on X pulled in more than 100 million views within days and did something a decade of hearings, forums, and vetoed bills hadn't managed. It got three senators from opposite ends of the AI debate onto the same piece of legislation.

Coxon left Anthropic this month following three years of work in pre-training research at Anthropic and OpenAI. He was not the first person from inside to ring this bell. Geoffery Hinton, Jan Leike, and Ilya Sutskever had all done this since 2023.

Congress ignored them all. Coxon’s ringing of the bell made some difference, however, because he had a colleague, Evan Hubinger, who held the position of alignment lead at Anthropic, confirm his allegations within hours.

Hubinger didn't hedge. He put the odds of AI-driven human extinction above 10% within a decade and said Anthropic has no concrete plan for controlling systems smarter than the people building them.

Two more Anthropic researchers echoed the concern. For a company that markets itself on safety, that kind of public admission from inside is hard for lawmakers to wave off as activist noise.

Congress has done this before time and again. Sam Altman requested regulation in 2023 and received no text at all in response. Schumer’s private “AI Insight Forums” gathered more than 60 senators but only created a framework, no law.

California’s SB 1047 went through the state legislature and died in Newsom’s office. Over 150 AI bills made their way through just one Congress recently; not a single one became law.

Sen. Ted Cruz, who was working throughout the past year to prevent states from regulating AI and lost that fight by a margin of 99-1, now finds himself as the co-sponsor of legislation alongside Sen.. Amy Klobuchar and John Thune. The new bill allows federal regulators to demand safety testing and prevent the deployment of any model that creates a catastrophic risk, such as bioweapons and nuclear threats.