Former Anthropic Researcher Warns AI Labs Are 'Gambling With Our Lives'
Jacob Coxon's resignation from Anthropic highlights growing insider fears over the race toward self-improving superintelligence.
A former AI researcher has sparked renewed alarm over the trajectory of artificial intelligence, claiming that leading labs are prioritizing speed over safety in a dangerous pursuit of superintelligence. Jacob Coxon, who resigned from Anthropic on September 9, 2026, warns that the current industry race is creating existential risks for humanity.
Coxon spent three years conducting pretraining research at both OpenAI and Anthropic. He stated that AI companies are not acting responsibly, specifically cautioning that these organizations are "racing straight to self-improving superintelligence and gambling with our lives." His departure follows a period of intense development at the world's most prominent frontier labs, where the goal of creating models capable of recursive self-improvement has become a primary objective.
The Push for Superintelligence
The debate over AI safety has intensified as frontier labs like OpenAI and Anthropic push toward increasingly powerful models. The concept of self-improving AI—where a system can rewrite its own code or design its own successors—is a central point of contention. While proponents argue this could lead to unprecedented breakthroughs in science and medicine, critics like Coxon suggest that the competitive pressure to achieve this milestone is overriding essential safety protocols. This internal friction reflects a broader divide within the AI community between those pursuing "accelerationism" and those advocating for a slower, more cautious approach to deployment.
Industry Implications
Coxon's warnings signal a shift in the AI safety discourse, moving from theoretical academic debates to urgent warnings from practitioners with direct experience in pretraining. If these concerns gain wider traction, they could lead to increased pressure for external oversight and the implementation of strict technical guardrails. Such a shift would fundamentally alter the operational model of AI labs, potentially slowing the pace of development in favor of national security and human safety. The risk, as framed by insiders, is that the window to implement these safeguards may be closing as models become more autonomous.
The Path Forward
As the industry continues to scale, the focus now turns to whether voluntary safety commitments from AI labs are sufficient to prevent catastrophic outcomes. While some policymakers have expressed interest in regulating the sector, the specific legislative mechanisms to monitor "self-improving" capabilities remain undefined. Observers are watching to see if other high-level researchers will follow Coxon's lead, which could trigger a larger exodus of safety-minded talent from the industry's leading firms.