The Fed rang the alarm about Anthropic's Mythos AI model — but had to go months without it

The Federal Reserve finds itself in a strange position regarding advanced artificial intelligence. While government officials warned top bank CEOs in April about the security risks posed by the Mythos AI model from Anthropic, the central bank itself lacked access to the software for months. This model, central to an initiative known as Project Glasswing, helps identify software vulnerabilities and security weaknesses.

Despite its status as a systemically important institution, the Fed spent months attempting to secure usage of the technology. Current Federal Reserve Chairman Kevin Warsh recently testified before the Senate, confirming that he is actively working to obtain access to Mythos and other cutting-edge models. Warsh emphasized that his concern extends beyond one specific tool, stating that the banking system needs to address potential security gaps as these models gain broader adoption.

Industry observers express surprise at this oversight. Given the Fed's role in setting financial policy, many expected it to occupy a primary position for evaluating such technology. Meanwhile, the broader landscape for AI development remains turbulent. The Trump administration has exercised significant oversight, including export controls that briefly hindered access to Anthropic's models before those restrictions were removed.

Competition is also mounting from abroad. Chinese startups like Moonshot AI continue to release models that challenge American capabilities, prompting concern among government officials and tech leaders regarding the current pace of domestic innovation. As the Fed continues its efforts to secure access, the delay highlights the friction between rapid technological advancement and the institutional processes required to manage national security and financial stability.