- Jacob Coxon claims Chinese spies have infiltrated US AI labs OpenAI and Anthropic
- He warns insider access enables covert theft, threatening US technological AI lead
- Coxon urges strict background checks and defense-level security at AI firms, unlike current minimal vetting
Former Anthropic and OpenAI researcher Jacob Coxon believes that Chinese spies have already infiltrated US AI frontier labs, OpenAI and Anthropic.
He made the explosive revelation to American entrepreneur Patrick Bet-David during a podcast. When Bet-David pushed him to reveal how sure Coxon was of this fact, he said "90% sure".
Coxon who became famous - after quitting Anthropic and going public with his concerns of rapid AI development at US frontier labs threatening the very existence of humanity - warns that insider access allows Beijing to conduct covert exfiltration and model theft at any time, neutralizing any technological lead Washington builds.
Coxon worked as a researcher at OpenAI and Anthropic sequentially, leaving OpenAI to join Anthropic, and then resigning from Anthropic in September.
Coxon's claim lands in the backdrop of US President Donald Trump warning against AI regulation, arguing that slowing down development hands an immediate victory to Beijing. "Whoever wins AI, wins"," Trump has said time and again.
While the White House treats raw speed as the ultimate defense against China, insider threats mean that accelerating development without strict counter-espionage measures simply accelerates what can be stolen.
Infiltration directly feeds into existing industry tensions over model distillation. US labs such as Anthropic and even government representatives have accused China of distilling US frontier models to release open-weight models.
AI distillation, also known as knowledge distillation, is a technique in which a large, complex and computationally expensive model, known as the "teacher", trains a smaller and faster model.
Coxon points out that unlike defense contractors like SpaceXAI, that face strict background checks and hiring restrictions regarding foreign nationals, Silicon Valley AI labs operate with minimal security vetting. AI firms have prioritised rapid recruitment of global talent over defense-grade counter-intelligence protocols, according to him.
Coxon advocates for mandatory background checks, nationalisation-style security protocols, and potentially restrictive hiring rules similar to aerospace defense standards.
In the meantime, the AI race between US and China rages on.
According to a Nikkei tabulation, 10 of China's leading AI companies - including DeepSeek, Alibaba/Qwen, Moonshot AI, Zhipu/GLM, Tencent, Xiaomi - launched 16 models in September.
Major US frontier labs such as OpenAI, Anthropic, Google, SpaceXAI, and Meta launched or updated at least 10-12 models, variants, and specialized reasoning sub-tiers during the same month.
Also Read: Despite Warnings Global AI Development Is Racing On
