Senior Anthropic Safety Researcher Resigns With Public AI-Risk Warning

A researcher who had worked on AI safety at both OpenAI and Anthropic resigned publicly this week, warning that AI companies are moving too fast and 'gambling with our lives' in the race to build ever more capable systems.

Key Points:

• Jacob Coxon, who had spent roughly four months at Anthropic after time at OpenAI, resigned on September 8 and said in a public message that leading AI companies have not fully internalized the risks of the technology they are building.

• Evan Hubinger, who leads Anthropic's Alignment Stress-Testing team, publicly stated his own estimate that there is roughly a 10% chance advanced AI causes human civilization not to survive the next decade.

• The resignation drew wide press coverage, including national television interviews, days ahead of Anthropic's widely anticipated public offering.

Public resignations with safety warnings are rare and carry outsized weight because they come from insiders rather than outside critics. Regardless of where one lands on the underlying risk estimates, the willingness of a company's own senior safety staff to say this publicly — during a high-stakes period for the company — is itself a governance signal worth watching.

Why It Matters: A rare public break from inside a leading safety-focused lab, days ahead of its expected public offering — boards and risk committees should treat public safety dissent from frontier labs as a governance signal, not just a media story.