Fed faced months-long access gap to Anthropic's Mythos AI after warning banks

The U.S. Federal Reserve experienced a significant delay in accessing Anthropic's advanced AI model, Claude Mythos Preview. For at least three months, the central bank did not have access to the model, even after convening a meeting in April to warn the nation's top banks about its potential cybersecurity risks.

The model, developed by AI firm Anthropic, is designed to excel at identifying software vulnerabilities. Anthropic had offered the model to a select group of banks and institutions as part of its Project Glasswing cybersecurity initiative. During this time, while other institutions began to patch identified weaknesses, the Fed itself was excluded.

Fed Chair Kevin Warsh confirmed the situation in congressional testimony in July. He stated that the central bank was still working to secure access to the Mythos model and other cutting-edge AI systems to better protect against potential threats. Warsh emphasized the necessity of access to such models for ensuring the security of the banking system and the Federal Reserve itself.

It remains unclear whether the Fed has since gained access to the Mythos model. The Fed declined to comment on the matter, and Anthropic did not provide a direct response to inquiries. Despite warnings about cybersecurity risks, the central bank faced challenges in obtaining a tool that could potentially aid in mitigating those risks.