Former Anthropic Researcher Jacob Coxon Resigns Citing 10% AI Extinction Risk by 2030, Pushes Non-Proliferation Treaty

Coxon’s exit from Anthropic followed three years at OpenAI and Anthropic. Within hours the claim reached every major outlet and prompted endorsements from Geoffrey Hinton on BBC Newsnight and additional researchers on the Joe Rogan podcast. The timing aligned with renewed calls for an AI non-proliferation treaty modeled on nuclear agreements. No primary technical paper or internal Anthropic document accompanied the resignation announcement.

The narrative serves two state-level interests. The United States can justify export controls and compute restrictions on China by framing AI as an existential risk requiring coordinated governance. China can accept limited transparency measures that slow private Western labs while preserving state-directed programs. Both sides gain leverage over domestic firms under the banner of safety without conceding core capability races.

Primary records show Hinton previously warned of job displacement rather than near-term extinction and that OpenAI and Anthropic have published scaling laws focused on capability gains, not termination probabilities. The Coxon thread on X had no prior activity before the coordinated rollout. No government has tabled a draft treaty text.

Next steps will likely include draft legislation in the US Congress and EU referencing the 10 percent figure, followed by Chinese counter-proposals that tie any limits to existing US chip sanctions. Regulatory momentum will depend on whether mid-2027 capability benchmarks exceed or fall short of current scaling projections.