U.S. financial leaders are sounding the alarm over a new generation of artificial intelligence that could dramatically escalate cyber threats to the banking system.

Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell recently convened top Wall Street executives in Washington for a closed-door meeting focused on emerging risks tied to an advanced AI model developed by Anthropic, according to a Bloomberg report.

The discussions centered on ensuring major banks are prepared for potential vulnerabilities linked to the technology, known as “Mythos,” and similar next-generation systems.

Officials are increasingly concerned that these tools could enable more sophisticated cyberattacks capable of targeting critical financial infrastructure.

Regulators view the threat as potentially systemic, given that the institutions involved are among the largest and most interconnected in the global financial system.

Powell’s participation underscored the seriousness of the issue, signaling that the risks extend beyond routine cybersecurity concerns.

Kevin Hassett, director of the National Economic Council, said policymakers are moving cautiously, emphasizing that safeguards must be in place before broader deployment of such powerful AI systems.

The concern stems from the model’s reported ability to identify — and in some cases exploit — hidden weaknesses across major software platforms and web browsers.

During internal testing, Anthropic indicated the system could autonomously discover ways to extract sensitive data across multiple environments.

That capability raises the prospect of faster, less predictable cyberattacks, where AI could be used offensively as well as defensively — accelerating what some experts describe as a new phase in the digital arms race.

Authorities outside the U.S. are also taking notice. Canadian officials and major banks have already begun reviewing potential risks, while U.K. regulators are preparing similar discussions with financial firms.

At the same time, policymakers are weighing how to incorporate these emerging threats into bank oversight. In recent years, regulators have required institutions to hold capital against operational risks, including cyberattacks — a framework that may need to evolve as AI-driven threats become more advanced and harder to quantify.

Anthropic has said it is limiting access to the technology and has engaged with government officials about its potential uses and risks.

Still, the rapid pace of development is forcing regulators and financial institutions to confront a new reality: AI may not just defend against cyber threats — it could also significantly amplify them.