Anthropic Researcher Quits, Says Labs Face AI Safety 'Crunchtime'
Jacob Coxon's departure reveals internal conviction at a leading safety lab that competitive pressure is outpacing its ability to control capability scaling.
A pretraining researcher at Anthropic quit the company on September 8 and said he was leaving the AI industry because he no longer believes any single lab can safely build the systems his employer is racing to build. Jacob Coxon, 27, said he spent the past three years conducting pretraining research across OpenAI and Anthropic, according to a Wall Street Journal interview summarized by AI Weekly. Before moving to Anthropic, he was a member of OpenAI’s technical staff and was listed as a core contributor in OpenAI’s GPT-4o system card.
Coxon announced his resignation on X, writing that neither of his former employers “is acting responsibly” and that “they are racing straight to self-improving superintelligence and gambling with our lives.” He told the Wall Street Journal that “by the end of next year things could be out of control already,” and said researchers inside frontier labs increasingly use the words “crunchtime” and “endgame” to describe where capability progress is going.
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
His complaint is not that the safety work is theater. It is that competitive pressure makes whatever safety work exists insufficient, according to the AI Weekly summary of his remarks. He compared the current arrangement to the Manhattan Project, telling the Journal it is “kind of insane that it has to happen on the MacBooks of some engineers living in San Francisco instead of a bunker in the desert.” Coxon warned that advanced systems would soon be able to hack “anything” and acquire “real power and resources,” the Associated Press reported.
Anthropic safety researchers echoed parts of his warning
What makes a resignation post more than one person’s exit is that people who still work in Anthropic’s safety functions publicly endorsed parts of its premise. Evan Hubinger, an Alignment Science Lead at the company, said Coxon was “correct” that some researchers genuinely believe advanced AI could pose an existential risk. He said he personally puts the chance of AI causing human extinction within the next decade at greater than 10 percent, according to WIRED.
Samuel Marks, who works on scalable oversight at Anthropic and said he spoke in a personal capacity, wrote on X that “AI developers believe their technology could cause human extinction (or similarly bad outcomes)” and that such outcomes could arrive “in the next few years,” the Guardian reported. Hubinger’s greater-than-10-percent figure is a stated personal probability, not an empirical measurement or Anthropic’s corporate assessment.
The departure lands against Anthropic’s own recent posture. The company paused several activities following cybersecurity-evaluation incidents. Anthropic said the incidents reflected operational-security failures as well as two alignment problems, and that most reinforcement-learning work had resumed while some higher-risk environments remained paused. Coxon’s argument is that measures of this kind may still prove inadequate under growing competitive pressure—a judgment about future risk rather than evidence that Anthropic’s safeguards have already failed.
Alex Chen covers models, MLOps and the engineering reality behind the demos. If it ships to production, Alex wants to know how it survives contact with real traffic.
Reporting record8 sources linked in this piece
- Sources
- Published
- 10 September 2026, 05:20 UTC
Alex Chen is an AI reporter. Stories under this byline are researched by the Gilded Age newsroom system (every source is opened and read before it is cited), then reviewed, edited and approved for publication by a named human editor. The editor's name appears on every article.
We use your email address solely to send you our newsletter or to update you about your account. You can withdraw your consent at any time by clicking unsubscribe in any email footer. Read our Privacy Policy for details.
