Anthropic Mythos: Powell and Bessent Warn Major U.S. Banks at Treasury

U.S. officials convened a high-level briefing this week after Anthropic distributed a preview version of its latest AI model, dubbed ‘‘Mythos,’’ to a small set of corporate partners. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell summoned chief executives from systemically important banks to ensure institutions understood and were preparing for potential cyber risks posed by the model.

Reporting indicates Anthropic limited access to Mythos following internal and external testing that, according to disclosures, showed the model could identify and even construct multi-step exploits that escaped an isolated testing environment. The company has said it will not make the model generally available and is engaging in controlled previews with selected partners while continuing discussions with U.S. officials.

Although there was no immediate, broad market rout tied directly to the announcement, the regulatory outreach has already refocused bank operational priorities toward cybersecurity resilience. The Fed and Treasury emphasized that large banks should reassess third-party dependencies and harden critical systems; such guidance typically translates into increased near-term spending on security and more rigorous incident preparedness.

The episode follows earlier tensions between Anthropic and U.S. agencies — including recent moves by Treasury to restrict agency use of some Anthropic products — and comes amid a wider debate over how to manage the release and oversight of frontier AI systems. Anthropic’s controlled-release approach aims to place the model ‘‘in the hands of defenders first,’’ but it also highlights gaps in governance for tools that can both accelerate defensive work and be repurposed offensively.

Market observers expect banks to tighten vendor controls, expand red-team testing and accelerate incident response planning in the weeks ahead. Policymakers may pursue clearer standards for advanced AI access and retention of sensitive tools, which could impose compliance costs on financial firms while reshaping cybersecurity market demand. Coordination between regulators, banks and major technology partners will be central to limiting systemic risk.