AI researcher quits, warns that superintelligence decisions are made on engineers' laptops

Jacob Coxon, a 27-year-old former Anthropic researcher, quit the AI industry, posting on X that pivotal decisions about superintelligent AI are being made inside a private Slack channel on engineers' MacBooks in San Francisco. He highlighted that both OpenAI and Anthropic are racing toward self-improving systems while senior staff privately fear the technology could wipe out humanity by decade's end. Evan Hubinger, who leads alignment stress testing at Anthropic, agreed, placing the odds of a fatal outcome above 10 percent and noting the lack of a concrete alignment plan.

The resignation prompted two more departures—Joe Benton and Josh Engels—who joined the AI risk nonprofit METR, and added to a series of exits by safety researchers from Anthropic and Google. In response, Anthropic CEO Dario Amodei published a 3,800-word essay proposing embedded third-party evaluators, coordinated democratic oversight, and eventual agreements with China, receiving public support from Sam Altman, Elon Musk and Demis Hassabis. The episode underscores growing internal alarm over AI safety and the absence of regulatory frameworks in the United States.