Anthropic researcher says humanity's future decided on 'MacBooks of engineers'

Anthropic researcher Jacob Coxon has left the artificial intelligence industry, claiming that the decisions shaping humanity's future are being decided in private company Slack channels. Coxon, a 27-year-old Briton with a background in mathematics, had spent three years at OpenAI and then joined Anthropic in 2026 due to its safety reputation. In a thread on X, he expressed concern that critical AI decisions are being made on MacBook computers owned by engineers in San Francisco, rather than in secure facilities.

Coxon noted that Anthropic staff debate the capabilities of their models internally via Slack, and that it's concerning to have such consequential choices made on personal computers. He warned that both OpenAI and Anthropic are racing towards superintelligence, gambling with human lives. Anthropic employees privately express fear about the potential for AI to kill everyone by the end of the decade, but soften their language in public.

Coxon highlighted a stark divide between OpenAI and Anthropic: while OpenAI appears to lack a clear sense of the societal stakes, Anthropic believes it must prioritize speed to avoid competitors acting irresponsibly. Evan Hubinger, who leads alignment stress testing at Anthropic, echoed Coxon's concerns, saying the company believes AI could kill all humans with a probability above 10% within the next decade, and admits the company does not yet have a plan to align superintelligence.

Other Anthropic researchers have followed suit and resigned, including Evan Hubinger, Joe Benton, and Josh Engels. Benton and Engels recently joined METR, a nonprofit evaluating catastrophic AI risk. Anthropic's chief scientist Jan Leike also left in 2024, citing a deterioration in safety culture. The company's CEO, Dario Amodei, responded by proposing a series of measures to slow AI development, such as embedding third-party evaluators with access to employee-like privileges and coordinating among labs in democracies.

These steps come as Anthropic prepares to go public, with responsible development touted as a key selling point.