SAN FRANCISCO - Anthropic researcher Jacob Coxon has resigned, warning that competition between leading AI labs could push development toward self-improving systems before adequate safeguards are in place.

Coxon announced his departure in a seven-post thread on X on Sept. 8 U.S. Pacific time, saying he had spent the previous three years doing pretraining research across OpenAI and Anthropic. “Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.”

Coxon later told Axios he quit two months before his Anthropic equity would have vested. “I left before any of my equity vested.”

Axios reported that Coxon still holds equity in OpenAI.

In his X thread, Coxon said people building advanced AI genuinely believe it could “kill us all by the end of the decade,” while executives and senior researchers often express those fears more cautiously in public. “This is not a marketing stunt.”

Coxon drew a distinction between his former employers. He said many people at OpenAI had not fully internalized what he called the stakes, while Anthropic understood them more clearly but remained trapped by competition with other developers.

In a subsequent interview with WIRED, Coxon described Anthropic as substantially more responsible than OpenAI and said it was not currently cutting corners. He predicted, however, that competition could eventually force trade-offs between rigor and safety.

Coxon proposed that OpenAI and Anthropic coordinate on limiting recursive self-improvement, in which AI systems are used to help develop more capable AI systems. He said broader coordination, including between the United States and China, would eventually be necessary.

His thread also raised the possibility of a temporary ban on improving model capabilities if needed to prevent a global race. In his final post, Coxon challenged researchers to consider whether they would launch a superintelligent reinforcement-learning run without a rigorous understanding of the system’s mind, or push for different conditions.

Coxon’s warning was echoed by Anthropic Alignment Science lead Evan Hubinger, who wrote on X that he personally puts the chance of AI killing all humans within the next decade at greater than 10%. In a follow-up, Hubinger said the risk from present models is low and that his concern is superintelligence arising through recursive self-improvement.

OpenAI did not return WIRED’s request for comment. Anthropic told WIRED that the industry would benefit from “a lawful, verifiable way to work together to pace how we release powerful models.”