US summons bank bosses over cyber risks from Anthropic’s latest AI model

The meeting

It has been reported that the US Treasury summoned chief executives of major banks to a meeting in Washington this week to discuss cyber risks tied to Anthropic’s new Claude Mythos model. Federal Reserve chair Jerome Powell was reportedly in the room. The guest list allegedly focused on leaders of systemically important banks — David Solomon, Brian Moynihan, Jane Fraser, Ted Pick and Charlie Scharf were said to have attended, while Jamie Dimon was invited but unable to make it. Not exactly small talk.

The Anthropic threat

Anthropic says its unreleased Mythos model has exposed thousands of vulnerabilities in widely used software and apps — some reportedly up to 27 years old — and that AI can now outpace “all but the most skilled humans” at finding and exploiting flaws. It has been reported that a leak of Claude’s code prompted the company to throttle the model’s release, limiting access to a handful of tech giants and infrastructure firms. Allegedly, the findings raise the prospect of attackers using AI to speed up password-cracking, encryption-busting and other high-impact intrusions.

Stakes and reactions

Banks and regulators have been tight-lipped; the Federal Reserve, Anthropic and several banks declined comment, and it has been reported that the Treasury did not respond to media requests. The meeting follows a recent US designation of Anthropic as a supply-chain risk — a label the company is contesting in court — and comes as CEOs increasingly warn that AI will amplify cyber risk. Why the urgency? Because when tools this powerful turn adversarial, the fallout could ripple through finance, public safety and national security.

This story is still developing. Expect more scrutiny — and more questions — as regulators, banks and tech firms try to square rapid AI advances with the brittle, century-old infrastructure that keeps money moving. Who’s ready to hold the line? Not many, it seems.

Sources: theguardian.com, Hacker News