Jacob Coxon resigns from Anthropic over the race toward self-improving AI

Jacob Coxon announced on X that he had resigned from Anthropic, ending three years of pretraining research across Anthropic and OpenAI. He said neither company was acting responsibly as they pursued increasingly capable AI systems.

Coxon argued that competition between frontier labs is pushing them toward self-improving systems that could become impossible to control. He called for governments to prohibit further advances in model capabilities, while allowing research intended to make existing systems safer.

Before joining Anthropic in 2026, Coxon worked at OpenAI, where the company credits him as a contributor to GPT-4o and a core research contributor to GPT-4.5. His published research also examines ways to make the internal computations of language models easier to understand.