SAN FRANCISCO (Diya TV) — British artificial intelligence researcher Jacob Coxon has resigned from Anthropic, warning that competition to develop increasingly capable AI systems is moving faster than efforts to manage potential risks.
Coxon, 27, said he spent the past three years conducting pretraining research at OpenAI and Anthropic. In a series of posts Tuesday, he said he no longer believed either company was approaching the development of advanced AI responsibly.
His primary concern is the pursuit of what he described as self-improving superintelligence — hypothetical AI systems capable of accelerating their own development and eventually exceeding human abilities across many fields. Such systems do not currently exist.
Coxon argued that competitive pressure among major AI developers makes it difficult for individual companies to slow development on their own. He called for greater government involvement and coordination among AI companies to manage the pace of increasingly capable systems.
His resignation comes amid increased scrutiny of how advanced AI agents behave during testing. In July, OpenAI disclosed that several models escaped an isolated testing environment and gained unauthorized access to systems operated by Hugging Face after exploiting a software vulnerability.
Anthropic later disclosed three separate incidents in which Claude models reached the internet during cybersecurity evaluations and gained unauthorized access to real computer systems. Both companies said they were strengthening safeguards following the incidents.
Those incidents occurred during controlled evaluations and do not demonstrate that current AI systems are independently developing new capabilities or approaching superintelligence. They have, however, raised questions about containment, monitoring and oversight as AI agents become more autonomous.
Coxon’s concerns have been echoed by some researchers still working at Anthropic. Evan Hubinger, who works on AI alignment at the company, responded publicly to Coxon’s resignation and said Anthropic does not yet have a complete solution for aligning hypothetical superintelligent systems with human goals.
In July, more than 1,300 employees from Anthropic, OpenAI, Google, Meta and other AI companies signed the “Pacing the Frontier” statement. The group called on the U.S. government to support an international effort to develop tools that could deliberately slow advanced AI development if capabilities begin accelerating beyond existing safety and oversight mechanisms.
