Beijing increasingly views Anthropic's Mythos — an AI model that hunts zero-day security vulnerabilities, restricted to vetted partners — as a potential offensive cyber weapon. Bloomberg reports that Chinese officials are growing worried the model could be wielded against China's networks, and are questioning why Anthropic denies Beijing access even for benign use.
Fudan University scholar Wu Xinbo framed the concern as rising to the level of national security. The anxiety is landing at a delicate moment: Xi Jinping is due to visit Washington on September 24, and both governments are preparing their first official AI talks under President Trump, expected in September. Beijing is reportedly weighing retaliatory measures if access to Mythos and comparable US models remains blocked.
The fear is about capability, not intent. Anthropic disclosed in April that Mythos can find and exploit software vulnerabilities, and US authorities had temporarily blocked foreign nationals from accessing the model. Analysts note China's own labs have already matched some of that capability — a reset in the cybersecurity AI race that cuts both ways and complicates any effort to control who gets the most powerful models.
Mythos has thus become a flashpoint in AI diplomacy: a symbol of Washington's refusal to share its most powerful systems, and of Beijing's determination not to depend on US technology. Whether September's talks can defuse the tension — or whether the AI rivalry turns into a full-blown cybersecurity confrontation — is now one of the defining questions of the US-China relationship.
