Top US and Canadian officials are sounding the alarm over a new AI model's cyber risks

This week, top financial regulators in both the United States and Canada pulled together emergency meetings with major banks to address a new kind of threat-not from hackers or foreign states, but from a cutting-edge artificial intelligence system. The focus: Anthropic's recently launched AI model, Mythos, which officials worry could be weaponized to launch highly sophisticated cyberattacks on financial infrastructure.

In Washington, Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell brought together executives from the country's largest banks for closed-door discussions about potential vulnerabilities. The goal was to assess how tools like Mythos-designed to reason, plan, and generate complex code-might be misused by bad actors to find security flaws, automate phishing campaigns, or even simulate large-scale financial disruptions.

Across the border, the Bank of Canada held its own session Friday with senior leaders from Canada's biggest financial institutions. While no incidents have been reported, officials stressed the need for vigilance. The concern isn't that Mythos itself is dangerous, but that its capabilities could lower the barrier for cybercriminals or state-sponsored groups to carry out high-impact attacks with minimal technical expertise.

Anthropic, the AI company behind Mythos, has emphasized that the model includes robust safeguards and is not publicly available in its full form. Still, regulators are acting on the principle that waiting for a breach could be too late. The meetings mark one of the first times financial authorities have proactively coordinated at this level over an AI model before any real-world harm has occurred.

What makes Mythos stand out is its ability to understand and manipulate complex systems-skills that are useful for researchers but could also be repurposed to map out weaknesses in banking networks or design malware that evolves in real time. While current AI models aren't fully autonomous threats, their speed and precision are raising new questions about defense readiness.

For banks, the message is clear: prepare for a new era of cyber risk where the tools of attack are smarter and more accessible. Some institutions are already increasing their red-team exercises, simulating AI-powered attacks to test their defenses. Others are working more closely with AI developers to understand the limits and risks of emerging models.

These conversations aren't about halting progress, but about staying ahead of it. As AI systems grow more capable, the line between innovation and risk is blurring-and regulators are realizing they can't afford to play catch-up.