What changed
Former OpenAI and Anthropic researcher Jacob Coxon says Chinese spies have infiltrated both US AI companies. Speaking on a podcast with entrepreneur Patrick Bet-David, Coxon said he was “90% sure” of the allegation and called for defense-level background checks and security controls.
Coxon worked at OpenAI before joining Anthropic, where he resigned in September. The report provides no independent evidence confirming infiltration and identifies no alleged spies.
Why it matters
The claim shifts the AI security debate from stolen model outputs to the people who can reach the systems behind them. If insider access exists, Coxon argues, Beijing could extract models or technical knowledge without waiting for a public release. That would make the race to build more powerful systems partly a race to protect what has already been built.
This lands alongside existing accusations that Chinese companies have used model distillation: a smaller “student” model learns from a larger “teacher” model. The distinction matters because distillation can be legitimate in principle, while covert access would turn it into a security breach.
Coxon’s proposed response would also change how these companies hire. Silicon Valley labs have recruited global talent at speed, while defense firms use tighter screening and restrictions on some foreign-national hires. Applying similar rules could protect sensitive systems, but it could also narrow the talent pool at a moment when development is accelerating on both sides of the Pacific.
The scale of that competition is clear in the figures cited in the report: Chinese companies launched 16 models in September, while major US labs launched or updated at least 10 to 12. Whether the allegation leads to investigations or policy changes at OpenAI and Anthropic remains unknown.
Comments
No comments yet.