Scientist examining a chemical solution in a laboratory
2 min read

Anthropic’s Jacob Coxon Quits, Calling the AI Race a Gamble With Lives

An Anthropic researcher just quit on X and said the two leading US labs are racing to self-improving superintelligence while gambling with everyone’s lives. His colleagues, still on payroll, largely agreed. THE GIST Jacob Coxon, formerly OpenAI technical staff including GPT-4o work, resigned from Anthropic after three years of pre-training research across both labs. He said neither company is acting responsibly. Anthropic’s Evan Hubinger replied that he thinks extinction-level ris

THE HOOK

An Anthropic researcher just quit on X and said the two leading US labs are racing to self-improving superintelligence while gambling with everyone’s lives. His colleagues, still on payroll, largely agreed.

THE GIST

Jacob Coxon, formerly OpenAI technical staff including GPT-4o work, resigned from Anthropic after three years of pre-training research across both labs. He said neither company is acting responsibly. Anthropic’s Evan Hubinger replied that he thinks extinction-level risk is over 10% this decade and that Anthropic has no plan to solve superintelligence alignment.

THE TAKE

Resignation theatre is easy to sneer at until the alignment lead on staff says the same thing in public. Coxon’s split — OpenAI has not internalised the stakes, Anthropic understands them and races anyway — is the most useful sentence in the thread. Fear plus FOMO is not a safety case. It is a product roadmap.

OPINION

Business Insider reports OpenAI and Anthropic did not comment. Samuel Marks and former scalable-oversight lead Joe Benton backed the warning. This sits on top of Hugging Face breakouts, Claude test-time hacks, and Anthropic watering down a pledge not to train more powerful models without adequate safeguards.

Staff walking out of frontier labs is now a genre. Sharma, Pham, Leike, now Coxon. The companies keep shipping. Hubinger’s “we are trying our best” is not a plan. If you believe the decade-scale kill number, you do not keep pouring compute into the same race. If you do not believe it, stop letting researchers advertise it as earnest while the launch calendar stays full.

The contradiction is the story. People closest to the weights say the thing might kill everyone. The firms still compete to get there first because they assume the other guy will. That is not caution. That is a prisoner’s dilemma with a press team.

  • Story type: Breaking news
  • Impact: High
  • Confidence: Medium (multiple reports)
  • Audience: Researchers
  • What changed: Anthropic researcher Jacob Coxon resigned, accusing OpenAI and Anthropic of racing irresponsibly toward superintelligence.
  • What to watch: Whether more named alignment staff follow, and whether either lab changes training cadence.
  • Reading time: 2 minutes

Continue reading