New Anthropic model forced the head of the US Treasury to hold an emergency meeting with top bank executives

According to sources familiar with the situation, the emergency meeting with US Secretary of the Treasury Scott Bessent and Federal Reserve Chairman Jay Powell was attended by the heads of leading American banks: Bank of America, Citigroup, Goldman Sachs, Morgan Stanley, and Wells Fargo. The top executives were already in Washington for an event hosted by a banking lobby group. JPMorgan Chase CEO Jamie Dimon also received an invitation but was unable to attend.

The reason for the meeting was the Trump administration's deep concern about the capabilities of a new artificial intelligence model from Anthropic. This system, named Claude Mythos Preview, demonstrated exceptional abilities in identifying cybersecurity vulnerabilities that could be exploited by malicious actors to attack the financial system.

Claude Mythos Preview was unveiled on April 7 to a limited circle of partners, including Amazon, Apple, and Microsoft. This was done to give them "an opportunity to prepare in advance to eliminate vulnerabilities."

Mythos is a "general purpose" model whose capabilities extend beyond cybersecurity. This is the first time Anthropic has deliberately limited the scale of a new product launch. In an official statement, the company noted that modern AI models have reached such a level in programming that in finding and exploiting vulnerabilities, they are capable of surpassing "almost any human, except for the most highly qualified specialists."

According to developers' data, Mythos has already identified thousands of serious flaws, including in "every major operating system and web browser." Some of them had gone unnoticed for decades.

Anthropic confirmed that it is consulting with the US government regarding the offensive and defensive capabilities of its system. At the same time, as FT notes, the limited release occurred against the backdrop of two incidents involving the leakage of Anthropic's own data, including the source code of its Claude assistant, which raised questions about the company's own security. The company explained these incidents as due to human error.

Currently, Anthropic, the Federal Reserve, and most of the banks that participated in the meeting have declined to provide official comments. The US Treasury Department and Citibank also did not respond to media inquiries.

FT observes that banking top executives have been warning about cyber risks to the financial system for years. In his annual letter, published this week, JPMorgan Chase's CEO noted that this "remains one of our greatest threats," and "AI will almost certainly make the situation worse," requiring significant investments in security.

• 9
• 1
• 3
• 1
• 1
• 0
• 1
• 0