The Trump administration is preparing to issue an executive order that will give federal agencies authority to review advanced AI models before public release, The New York Times reported Thursday, citing people familiar with the plan.

The order, which may come as early as today, will task the Office of the National Cyber Director with creating a formal evaluation process within two months to detect security vulnerabilities. Officials are also proposing that leading AI companies voluntarily share models between 14 and 90 days in advance.

Mythos raises debate over AI-driven cyber risk in finance systems

The anticipated order comes amid mounting concerns among global finance leaders over Anthropic’s Claude Mythos AI model, which experts warn could be capable of uncovering and exploiting cybersecurity weaknesses in critical financial and digital infrastructure systems.

Early testing has reportedly shown the model can identify vulnerabilities in operating systems, web browsers and enterprise software, raising fears of systemic risk if misused.

Last month, Treasury Secretary Scott Bessent and Federal Reserve Chair Jay Powell took the unusual step of calling in senior executives from major US banks for direct briefings on cyber risks linked to the AI model.

Canadian Finance Minister François-Philippe Champagne said the issue was discussed extensively at IMF meetings, calling it an “unknown unknown” that requires urgent coordination to safeguard financial system stability.

Anthropic has not released Mythos publicly, instead limiting access to partners including Microsoft, Amazon Web Services, Nvidia and CrowdStrike through a controlled security initiative.

Independent evaluations suggest the model is highly capable in insecure environments but not dramatically more advanced than previous versions, though governments and financial institutions are actively testing the system ahead of any wider deployment.

Major crypto exchanges like Coinbase and Binance are seeking early access to Mythos as the industry prepares for heightened AI-driven cybersecurity risks, according to The Information. Coinbase has been in close communication with Anthropic, while other firms are also assessing its potential impact on crypto security.