AI researcher Coxon, who quit Anthropic last month after accusing it and OpenAI of gambling with lives, said in a podcast that slowing US AI development may be ineffective because Chinese spies may have infiltrated US labs, according to Business Insider

Espionage claims of this kind sit in a well-worn genre: a named insider publicly quantifying penetration of strategically sensitive firms. On past occasions such allegations have mattered less as verified intelligence than as inputs to the policy debate, feeding export-control tightening, outbound investment screening, and security-clearance scrutiny of lab hiring, the channels through which they have historically reached listed AI-exposed equities rather than through any direct earnings effect. The argument itself, that diffusion of model progress makes unilateral US slowdown ineffective, mirrors a line previously advanced by industry and national-security voices lobbying against stricter development controls, and tends to surface when regulation is actively contested. The 90% figure is one individual's assertion in an interview, not a finding, and neither lab has confirmed any breach; denial or silence from the companies is the usual immediate sequence. Worth noting is whether official bodies or lawmakers pick up the claim, since that conversion from podcast remark to policy citation is what has historically given such stories durability. Absent that follow-on, single-source accusations of this type have tended to fade within a news cycle.