The White House’s restriction on Anthropic’s Mythos model is now being tied to concerns that a China-linked group may have accessed it. If confirmed, that would move the episode beyond ordinary product-risk management and into the realm of export controls and national security.
For AI labs and their customers, the practical lesson is that access decisions may be shaped by intelligence and geopolitical concerns as much as by benchmark performance. Frontier models are increasingly being treated as strategic infrastructure, not just software products.