'Gambling With Our Lives': Why Researcher Jacob Coxon Resigned From Anthropic AI

A 27-year-old pretraining researcher, Jacob Coxon, has resigned from Anthropic after three years across two top artificial intelligence laboratories, warning that tech firms are building self-improving superintelligence without adequate safety controls. Coxon, who previously worked at OpenAI, announced his exit on social media platform X, stating that neither company is acting responsibly as they race toward recursive artificial intelligence capabilities.

Warnings From Within

Coxon claimed that senior leaders across major AI firms hold deep private anxieties regarding existential risks that contrast sharply with their reassuring public statements. He described impending "superhuman systems" capable of hacking network security protocols, revolutionising fields overnight, and independently acquiring power, characterising the private development push as a dangerous corporate gamble.

A Systemic Race Dilemma

Detailing organizational cultures in Silicon Valley, Coxon argued that while OpenAI staff have failed to deeply internalise civilisational risks, Anthropic researchers understand the stakes yet feel trapped in a global race. Responding to the post, Anthropic Alignment Science Lead Evan Hubinger acknowledged Coxon's points, stating he personally calculates a greater than 10 per cent chance of existential catastrophe from AI within the decade due to rapid recursive self-improvement.

Policy Measures and Pause Calls

The fallout highlights growing friction between competitive tech expansion and international security considerations, even catching the attention of tech leaders like Elon Musk, who recently praised Anthropic's models despite ongoing industry rivalries. Coxon urged remaining researchers to challenge current development conditions, calling for coordinated industry measures, such as a temporary global ban on capability improvements, to prevent disaster.