Anthropic’s Claude Mythos Preview, a potent AI model that has alarmed cybersecurity experts and financial regulators on both sides of the Atlantic since its limited release earlier this month, is being made available to major federal agencies through the establishment of security protocols by the White House Office of Management and Budget. According to a memo that Bloomberg examined, Federal CIO Gregory Barbaccia told Cabinet departments about the idea.
The action is taken despite the ongoing legal and political dispute between the Trump administration and Anthropic regarding the usage of the company’s AI technology, as well as government officials’ concerns about the model’s potential to significantly raise cybersecurity risk.
A Model Too Risky For Public Distribution
On April 7, Anthropic unveiled Claude Mythos Preview, limiting access to about 40 partner organizations as part of Project Glasswing. According to a Sullivan & Cromwell analysis of the disclosure, the company revealed that during internal testing, engineers without formal security training instructed the model to find software vulnerabilities overnight and “woke up in the morning to discover not only that Mythos had done so, but that the model had also provided them with a complete, fully functional method of exploiting those vulnerabilities.”
Less than 1% of the thousands of previously undiscovered, high-severity vulnerabilities that Mythos had found across all major operating systems and web browsers have been patched, according to Anthropic. According to the corporation, there are no intentions to make the model widely accessible.
Despite tensions, Anthropic informed the Trump administration on Mythos, according to co-founder Jack Clark at the Semafor World Economy summit. “The government has to know about this stuff,” Clark stated.
Wall Street And Washington Scramble
On April 7 in Washington, Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell called an urgent conference with CEOs of large U.S. banks to alert them about the ramifications of the model, as first reported by Bloomberg. Since then, the model has been internally tested by JPMorgan Chase, Goldman Sachs, Citigroup, Bank of America, and other major banks.
According to Bloomberg, Treasury CIO Sam Corcos independently started looking for direct access to Mythos so the department’s IT team could look for weaknesses in its own systems. According to Politico, Anthropic has also been contacted by three congressional committees and at least two additional federal agencies to test the model.
European Authorities Participate In The Reaction
According to Bloomberg, the European Central Bank is currently holding a call with chief risk officers of eurozone lenders to evaluate Mythos’ concerns. Supervisors are trying to comprehend the model’s implications for legacy banking technology, and the ECB is collecting data on the model’s ability to exploit weaknesses in financial systems. According to Reuters, ECB supervisors see Mythos as a threat to the banking sector that has alarmed US and British regulators.
The swift, international regulatory reaction highlights a fundamental conflict: the same features that make Mythos an effective defensive tool for identifying and resolving software defects also make it a potential weapon if it ends up in the wrong hands.