Treasury, Fed push banks to test AI that Pentagon called a national security risk.

The Trump administration is pushing the country's largest banks to use an AI system that the Pentagon has formally designated as a national security risk.

Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell held a meeting with executives from Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley this week where they encouraged them to deploy Anthropic's Mythos model to hunt for vulnerabilities in their systems, TechCrunch reported. The four banks are already testing the model, TechCrunch noted.

Mythos, which Anthropic announced April 7, can identify and exploit previously undiscovered flaws across every major computer operating system and browser. In early testing it found thousands of zero-day vulnerabilities, including a flaw in FFmpeg that had gone unpatched for 16 years, Reuters reported. The Cloud Security Alliance says Mythos fundamentally changes the economics of vulnerability research by lowering the technical skill required to find and exploit security holes faster than organizations can patch them.

Yet the Defense Department has barred Anthropic from Defense Department systems. The Pentagon labeled Anthropic a supply chain risk after the company refused to let the Defense Department use its models for autonomous weapons and domestic surveillance, Nextgov reported. A federal contracting dispute remains active.

Jack Clark, Anthropic's co-founder, confirmed at a Semafor event Monday that the company is in active discussions with the Trump administration about Mythos and future models, Reuters reported. He argued that the government needs visibility into advanced AI systems. He also offered a striking concession: "This is not a special model," Clark said. "There will be other systems just like this in a few months from other companies, and then a year to a year and a half later, there will be open-weight models from China that have these capabilities," Nextgov reported.

Clark's framing essentially says: the capabilities are coming regardless, so Washington might as well have a relationship with the lab that built them. The 12-18 month window before China fields comparable open-weight systems is the time window where that relationship has strategic value.

But the harder question is what accountability structure exists when a company simultaneously fights the Pentagon in court over its AI use restrictions while the Treasury Department promotes its product to the financial sector. The banks running Mythos have no independent mechanism to audit what the model does with the vulnerabilities it finds or where that information flows. Anthropic argues capability overhang is a structural feature of frontier AI, not a defect the company can simply fix. That argument is convenient for a company seeking government partnership. It is less reassuring for the banks being asked to trust it with their security posture.

The UK financial regulator is also examining Mythos, TechCrunch noted, citing the Financial Times, suggesting this is not only an American arrangement. What remains to be seen is whether the Treasury's push to deploy Mythos across major banks represents a legitimate defensive measure or an acceleration of the dynamic Clark described as inevitable.

The Pentagon dispute is ongoing. The bank testing continues. At some point those two facts will force a single policy question: what leverage does the government actually have over a company whose model it has simultaneously classified as a supply chain risk and a critical defensive asset?