Former OpenAI researcher claims China has all US AI code, gets roasted by industry

Will Depue, a former OpenAI researcher, is making some wild claims. He says China has already fully obtained all of OpenAI and Anthropic's GitHub code repositories, Slack chat logs, and internal documents. He even speculated that he wouldn't be surprised to see "plausibly deniable" stolen architecture in Chinese open-source models.

When pressed for evidence, Depue argued his conclusions come from conversations with security researchers, and accused the industry of overestimating frontier labs' defenses. Depue, who frequently posts outlandish takes on social media, also once announced on April Fools' Day that he was joining DeepSeek.

Unsurprisingly, the AI community has been having a field day mocking these claims. AI scholar Nathan Lambert called it "self-serving nonsense" from frontier labs, noting that the real flow of technical know-how happens through Silicon Valley's talent mobility, bar hangouts, and lack of non-competes — not Slack logs. The real moat, he argued, is compute and engineering resources.

Researcher Elie Bakouch responded with a perfect parody, using the same format to claim that "without a doubt, all frontier labs today have completely mastered DeepSeek, Kimi, GLM, and MiniMax's technical reports and models" — a sharp jab at how US labs freely benefit from Chinese open-source work while wallowing in persecution fantasies.