U.S. warns of new AI-related risks to the banking system

The U.S. Treasury Department and the Federal Reserve held an extraordinary meeting with leaders of major banks to warn about new risks that artificial intelligence (AI) could pose to the stability of the financial system.

The meeting took place in Washington and included participation from leading financial groups such as Bank of America, Citigroup, Goldman Sachs, Morgan Stanley, and Wells Fargo—banks considered systemically important to the U.S. economy.

Source of the concern

Officials said the concerns stem from a new AI model named Mythos developed by the technology company Anthropic. The model can detect and exploit vulnerabilities in banks’ computer systems. If misused by bad actors, officials warned it could infiltrate payment systems, disrupt lending operations, or steal data from millions of customers.

This was described as a rare instance in which the U.S. Treasury and the Fed jointly summoned banks to issue a direct warning, reflecting the seriousness of the threat.

How the technology works and why it matters

Anthropic has also warned about the potential risks of Mythos. The model is currently limited to about 40 trusted partners, including major technology firms such as Microsoft, Apple, and Google, as well as some financial institutions.

The stated aim of limiting access is to allow these entities to test and fix security vulnerabilities before the technology is widely deployed.

During testing, Mythos demonstrated strong performance in identifying bugs in common operating systems and browsers, even outperforming earlier models. However, officials said the same capability could be misused for cyberattacks.

Additional reasons for restricting access

Beyond security concerns, restricting Mythos issuance is also intended to prevent copying of the technology and to address current data-processing infrastructure constraints.

Broader context and policy pressure

On the political side, Anthropic faces pressure from the U.S. government. It is classified as a supply-chain risk and faces restrictions from the Pentagon after refusing to allow use of its technology for certain military purposes.

Officials’ assessment and expected response

U.S. officials said AI risk is no longer hypothetical but a concrete threat to the financial system. They urged banks to quickly upgrade security measures and bolster defenses to help ensure the safety and stability of the broader system.