‘No Company Is Acting Responsibly’: Lead Anthropic Researcher Leaves Industry Over AI Control Fears
·
09.09.2026 12:25
·
2 min read
Researcher Jacob Coxon, formerly of OpenAI and head of model pre-training at Anthropic for the past three years, has announced his departure from the company and a complete exit from the artificial intelligence industry. Coxon stated that an aggressive commercial race is forcing leading labs to compromise safety protocols in a rush to build self-improving algorithms.
Compromises for Market Survival
In a public statement and an interview with The Wall Street Journal, the 27-year-old researcher detailed the reasons behind his resignation. Coxon initially joined Anthropic because of its reputation as a safety-first lab that prioritized ethics over commercial gains. In practice, however, the pressure to keep pace with OpenAI and other tech giants turned those principles into a mere formality.
“No company is acting responsibly right now. Inside the labs, many people genuinely believe catastrophic scenarios could unfold by the end of the decade, yet no one can afford to hit the brakes for fear of losing their lead,” Coxon said.
Coxon warned that the pace of advancement is outpacing the understanding of underlying architectures: modern neural networks can already autonomously discover software exploits and assist in code development, while reliable mathematical methods to control superhuman systems remain non-existent.
The Threat of Recursive Self-Improvement
His concerns directly align with a recent research note from the Anthropic research institute on recursive self-improvement (RSI). The paper acknowledges that when algorithms begin writing code at scale to train their next iterations, the risk of a critical loss of control increases dramatically.
As noted by Business Insider, Coxon’s departure follows a series of high-profile resignations across safety teams at top AI labs.
Coxon is urging governments to establish strict regulatory oversight of the industry and legally enforce a temporary moratorium on scaling frontier model parameters. While his exit does not prove that out-of-control superintelligence is already here, it highlights the growing alarm within top-tier AI research teams.
