Anthropic fights Pentagon blacklist as banks test new security model

Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell met behind closed doors with bank chiefs and urged them to test Anthropic’s newest security model against their own systems, even as the Pentagon has kept the company on a supply-chain blacklist. The split-screen moment laid bare a growing policy contradiction: Washington is warning about the cyber risk posed by advanced AI, while simultaneously asking major lenders to use the same technology to harden their defenses.

Anthropic said it received a letter from the Department of War on March 4 confirming that it had been designated a supply-chain risk to America’s national security. The company said on March 5 that it would challenge the move in court, arguing that months of negotiations broke down over Anthropic’s requests for exceptions involving mass domestic surveillance of Americans and fully autonomous weapons. A federal appeals court declined on April 8 to temporarily block the designation while the case moves forward, leaving the blacklist in place for now.

At the same time, Anthropic rolled out Claude Mythos Preview only in limited form, saying the model can identify and exploit zero-day vulnerabilities in every major operating system and every major web browser. The company said more than 40 organizations are involved in Project Glasswing or related defensive-security work, with launch partners including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks.

The bank meeting underscored how quickly Mythos has moved from a theoretical risk to a live policy issue in finance. Goldman Sachs, Citigroup, Bank of America and Morgan Stanley were already testing the model internally or expected to gain access soon. The executives present included Jane Fraser of Citigroup, Ted Pick of Morgan Stanley, Brian Moynihan of Bank of America, Charlie Scharf of Wells Fargo and David Solomon of Goldman Sachs. JPMorgan CEO Jamie Dimon was invited but did not attend.

The episode raises a larger question about AI governance: if federal security officials view Anthropic as a national-security risk, why are they also encouraging the country’s biggest banks to experiment with its most advanced cyber tool? The answer may be that the model’s offensive potential is exactly what makes it useful for defense. But the mixed signals also risk eroding trust in how Washington sets rules for frontier AI, especially when the same model can be treated as both a threat and a tool.