Fed sought access to Anthropic’s Mythos AI model amid cybersecurity concerns

The Federal Reserve and Treasury Department raised concerns in April about Anthropic’s Claude Mythos Preview, saying the model could help expose cybersecurity weaknesses at major financial institutions. The AI tool was offered to a limited set of banks and other organizations through a cybersecurity effort called Project Glasswing. According to the report, the Fed did not have access to the model for at least three months after the meeting, even as other participants used it to review vulnerabilities. Fed Chair Kevin Warsh later told senators that he was still working to secure access for the central bank and other institutions. Anthropic has expanded access to Project Glasswing in stages, while the U.S. government has also been adjusting AI policy and export controls. The article says the episode underscores how financial regulators and banks are trying to keep pace with fastchanging AI tools and the security risks they may create.