Powell and Bessent warn bank CEOs about risks from Anthropic's new AI model

Jerome Powell and Scott Bessent held closed-door discussions this week with chief executives from some of the nation's largest banks over a growing concern: the potential misuse of Anthropic's newly released AI model, Mythos. The talks, confirmed by multiple sources, reflect mounting unease among regulators about how quickly advanced artificial intelligence is outpacing current safeguards in the financial sector.

Anthropic, the AI company behind the Claude series, introduced Mythos to a limited group of enterprise clients earlier this week. While the model promises breakthroughs in automation and problem-solving, internal testing and early external analysis suggest it can generate highly convincing phishing content, exploit code, and even mimic human behavior in digital environments with alarming accuracy. That capability has raised red flags about its potential use in coordinated cyberattacks on financial institutions.

According to people familiar with the conversations, Powell and Bessent emphasized that the financial system's stability could be at risk if threat actors gain access to tools like Mythos. The discussions focused on how banks are managing AI exposure, whether through third-party vendors or internal development, and what contingency plans exist for AI-driven cyber incidents.

One major bank executive, who spoke on condition of anonymity, said the tone of the meeting was not alarmist but urgent. "This isn't science fiction anymore," they said. "We're talking about real tools that can automate attacks at scale. The Fed and Treasury want to make sure we're not caught flat-footed."

Anthropic has stated it implemented strict access controls and monitoring for Mythos, including usage audits and rate limits. But experts note that once AI models are deployed-even in restricted settings-there's always a risk of leakage, reverse engineering, or misuse by authorized users. Some cybersecurity firms have already detected early attempts to mimic Mythos-like behavior on underground forums.

The response from regulators marks a shift from general AI oversight to targeted, sector-specific risk management. Financial institutions are now being asked to report on their AI governance frameworks, with potential new guidelines expected within months. The goal is to balance innovation with resilience, especially as AI becomes embedded in trading, customer service, and fraud detection systems.

For now, no breaches tied to Mythos have been reported. But the fact that Powell and Bessent stepped in so quickly underscores how seriously they're taking the threat. As one official put it: "We're not waiting for the fire. We're checking the smoke detectors."