Anthropic’s latest AI model strikes fear into banks

 Anthropic’s latest AI model strikes fear into banks

Morning Brew Design

An impromptu meeting of bank CEOs and federal officials was held in the nation’s capital this week because of the destructive capabilities of Anthropic’s latest AI model, Claude Mythos, which can detect cybersecurity flaws in operating systems and web browsers with exponentially greater efficiency than human hackers. Fed Chair Jerome Powell and Treasury Secretary Scott Bessent gathered the heads of Bank of America, Citigroup, Goldman Sachs, Morgan Stanley, and Wells Fargo on Tuesday after the limited rollout of Mythos, per Bloomberg. The emergency meeting is a sign that the Trump administration thinks that, while Mythos is intended to protect companies from hackers, it could be used to attack the foundation of the US financial system by targeting the world’s biggest banks. What’s so scary? Anthropic distributed Mythos to only ~40 organizations, which include major banks, according to Bloomberg, because the company said it’s too powerful for a full release to the public. CNBC reported that the company also briefed government officials on its capabilities before the release. So, how dangerous is it? Officials believe Mythos can debilitate Fortune 100 companies, infiltrate national defense systems, and take down huge chunks of the internet, according to Axios. A security expert told Business Insider that a team of humans can discover about 100 critical flaws with no immediate fixes per year (not including what happens on bad first dates), but Mythos can find “thousands.” Offense vs. defense: Mythos is equally capable of identifying and exploiting weaknesses in a system. Experts told BI that this gives hackers an advantage in the short term. But as widespread adoption occurs, the edge shifts back to those defending themselves. Safety net: Regulators require banks to hold capital in reserve to cover unexpected losses from events such as data breaches and cyberattacks. But banks have complained that they require too much, and a proposal from the Fed last month, if approved, would ease some of those requirements. Zoom out: The concerns over Mythos come as Anthropic is fighting the Trump administration over being designated a supply chain risk by the Pentagon after limiting the use of its AI tech in war.