Anthropic’s Mythos model reportedly found vulnerabilities in classified U.S. government systems, according to an AP report cited by Reuters, in a story that lays out a familiar arrangement: private tech firms chasing IPO money, and the state deciding which tools it wants for surveillance and war. The company, which is preparing for an IPO, has clashed with the U.S. government over the use of its models, and the government responded by placing Anthropic on a national security blacklist, according to the report.
The State Wants the Machine
The report says Anthropic refused to allow the U.S. military to use its AI for domestic surveillance and fully autonomous weapons. That refusal is the point where the polite language of innovation runs into the blunt edge of state power. The military wanted the model for domestic surveillance and fully autonomous weapons; Anthropic said no. The U.S. government then placed Anthropic on a national security blacklist, according to the report. The sequence is tidy enough: a corporation with a model, a state with a demand, and a blacklist when the answer does not fit the security script.
Reuters said the relationship between Anthropic and the U.S. government has been rocky. That is the diplomatic version. The less polished version is that the state and the company are negotiating over who gets to control the machinery, and for what purpose. The report does not say the government abandoned its interest in the model. It says the company refused the military’s requested uses, and the government answered with exclusion.
IPO Pressure, Public Power
Anthropic is preparing for an IPO, which places the company in the usual orbit of capital discipline: growth, valuation, and the pressure to look indispensable. The report does not spell out the market mechanics, but it does show the company moving in the same ecosystem where private firms sell advanced systems and governments demand access to them. The result is not a clean line between public authority and private enterprise. It is a shared infrastructure of control, with the state holding the blacklist and the company holding the model.
The Reuters article, citing AP, says Anthropic’s Mythos model reportedly found vulnerabilities in classified U.S. government systems. That detail matters because it shows how deeply these systems are already entangled with the tools of private AI firms. Classified systems are not outside the reach of corporate technology; they are part of the same managed environment, where security is invoked loudly and vulnerability appears anyway.
Security for Whom
The government’s response was not to explain the vulnerabilities in any public way, at least not in the report. Instead, it placed Anthropic on a national security blacklist. That is how the state tends to speak when it wants control without debate: not through transparency, but through exclusion. The blacklist is a bureaucratic instrument, but it carries the force of a political decision about who may participate in the state’s security apparatus and who may not.
The report gives no sign of any democratic oversight over the military’s interest in domestic surveillance or fully autonomous weapons. It does, however, show the usual arrangement in which the public is told to trust institutions that are already shopping for AI systems while keeping the details out of sight. Anthropic’s refusal, the government’s blacklist, and the rocky relationship between the two are not a side story. They are the story: a state seeking machine power, a company seeking market power, and ordinary people left to live under both.