Anthropic Researcher Says AI Has More Than 10% Chance of Killing All Humans As Colleague Quits Over the Dangers

Evan Hubinger, an alignment science lead at Anthropic, made his assessment after Jacob Coxon announced his resignation from the company. Coxon argued that he was leaving because leading AI developers were racing straight to self-improving superintelligence and gambling with human lives without adequate safeguards. Hubinger agreed with his colleague and stated that he personally believes there is a greater than 10% chance AI could kill all humans within the next decade, adding that the company does not yet have a solution for aligning future superintelligent systems with human interests.

Alignment researchers worry that advanced AI systems might pursue objectives conflicting with human interests, become difficult to monitor, or circumvent safeguards. This challenge intensifies if AI systems can contribute to recursive self-improvement, creating a feedback loop where capability development outpaces human oversight. Although there is no established scientific consensus regarding human extinction, the core technical question remains whether humans can reliably constrain and correct systems that outperform them.

Coxon also warned that competition between companies and countries could make meaningful coordination on safety measures difficult. While companies like Anthropic and OpenAI invest billions of dollars to develop more capable models, safety researchers warn that protective mechanisms may not advance at the same speed as AI capabilities.