The Trump administration placed Anthropic on a blacklist, but now it is urging banks to utilize its AI.

      In summary: Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell are encouraging Wall Street's largest banks to evaluate Anthropic's Mythos AI model for cybersecurity weaknesses, despite the Pentagon's legal battle against Anthropic, which it labeled a supply chain risk for its refusal to eliminate safety precautions related to autonomous weapons and mass surveillance. Reportedly, JPMorgan Chase, Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley are all testing the model. Mythos has discovered thousands of zero-day vulnerabilities across major operating systems and browsers and is being made available through a limited program called Project Glasswing to around 50 organizations. UK regulators are also racing to assess the associated risks.

      The Trump administration is quietly pushing America’s largest banks to test the same AI technology it has been trying to undermine for the past two months. As reported by Bloomberg, Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened executives from JPMorgan Chase, Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley this week and urged them to utilize Anthropic’s new Mythos model to identify cybersecurity vulnerabilities in their systems.

      This recommendation is striking due to its contradictory nature. Currently, Anthropic is engaged in a court battle with the Department of Defense after Defense Secretary Pete Hegseth labeled the company a "supply chain risk," which prevents it from securing military contracts and instructs defense contractors to cease using its technology. This designation followed Anthropic's refusal to lift two safety measures on its AI models: prohibition on usage in fully autonomous weapons and restrictions against mass surveillance of U.S. citizens.

      Now, two of the administration’s top economic officials are advising Wall Street to adopt the very technology that the Pentagon is attempting to blacklist.

      What Mythos actually does

      The Claude Mythos Preview is a cutting-edge model that Anthropic did not specifically develop for cybersecurity. Its ability to identify vulnerabilities emerged as a byproduct of improvements in code reasoning and autonomous functioning. During testing, Mythos uncovered thousands of zero-day vulnerabilities, previously undiscovered flaws found in every significant operating system and web browser.

      The findings were significant enough that Anthropic decided against a public release of the model. Instead, it initiated Project Glasswing, a controlled program that provides access to approximately 50 organizations, including Amazon Web Services, Apple, Google, Microsoft, Nvidia, Cisco, CrowdStrike, Palo Alto Networks, and JPMorgan Chase. As part of this initiative, Anthropic has pledged up to $100 million in usage credits and $4 million in direct donations to open-source security organizations.

      The narrative that the model is "too dangerous to release" has drawn skepticism. Tom’s Hardware pointed out that assertions of “thousands” of severe zero-day discoveries were based on merely 198 manual reviews, and many of the identified vulnerabilities were in outdated software or very difficult to exploit. Some security experts suggested that the limited release resembled less a responsible approach to AI governance and more a clever sales strategy: create scarcity, induce fear, and attract customers.

      The Pentagon paradox

      The clash between the Bessent-Powell recommendation and the Hegseth designation symbolizes not just mixed signals, but two branches of the same administration pursuing openly contradictory policies concerning the same company.

      The Pentagon's conflict began in February when Hegseth gave Anthropic CEO Dario Amodei a Friday deadline to remove the company’s safety restrictions or risk losing its $200 million defense contract, a request Amodei declined. Hegseth subsequently categorised Anthropic as a supply chain risk, prompting President Trump to instruct federal agencies to cease using its technology. A Pentagon official even accused Amodei of having a "God complex." Trump labeled Anthropic as a "radical left, woke company."

      The courts have since taken divergent stances. A federal judge in California issued a preliminary injunction that blocked the supply chain designation, asserting that “nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the US for expressing disagreement with the government.” Conversely, an appeals court in Washington, D.C., denied Anthropic’s request to halt its blacklisting temporarily while the case progresses. The cumulative outcome: Anthropic is barred from DoD contracts but remains eligible to work with other government agencies.

      It is in this context, with Anthropic excluded from the Pentagon yet still operating with the Treasury and the Fed, that Bessent and Powell intervened this week.

      What the banks are actually doing

      Although JPMorgan Chase is the only bank listed as an official Project Glasswing partner, Bloomberg reported that Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley are also testing Mythos in-house. The reported applications include vulnerability detection, fraud-risk identification, and compliance workflow automation within financial systems.

      The quick adoption reflects a real concern. If Mythos can detect zero-day vulnerabilities in operating systems and web browsers, it can presumably do the same in