The White House's recent decision to slap export restrictions on Anthropic's cutting-edge AI model, Mythos, was reportedly fueled in part by fears that a China-linked group may have already gotten their hands on it. According to a Semafor report, U.S. officials are concerned that if the Chinese government truly accessed Mythos 5 or its sibling model Fable 5, it would represent a significant national security threat. One particular worry is distillation, a technique where a less powerful "student" AI is trained on a more advanced model to mimic its capabilities, potentially allowing adversaries to reverse-engineer the technology without building it from scratch.
The White House has not publicly confirmed these details, and the silence is notable. David Sacks, a tech advisor to President Trump, took to X to address the export controls but pointedly did not mention China. Instead, Sacks focused on reports that Mythos and Fable could be jailbroken, a claim Anthropic has pushed back on. A spokesperson for Anthropic told Semafor that China was never raised during the government's discussions about export restrictions, and the company did not respond to a request for comment from The Verge.
This would not be the first time Mythos has been exposed to unauthorized access. Anthropic has publicly stated that the model is too powerful and dangerous for general release, yet a Discord community reportedly held access to it for two full weeks before the company discovered the breach and shut it down. The repeated security concerns around a model that Anthropic itself considers too risky for public consumption are raising fresh questions about how the company is safeguarding its most sensitive AI systems. With the geopolitical stakes of frontier AI continuing to climb, the intersection of corporate security lapses and international espionage risks is becoming an increasingly uncomfortable spotlight for one of the industry's most closely watched players.