This week, US Treasury Secretary Scott Bessent brought together executives from leading American banks in Washington to examine cybersecurity threats linked to Anthropic’s newest AI system, according to sources. Participants reportedly included Federal Reserve Chair Jerome Powell, who met at the Treasury’s main office following the introduction of the Claude Mythos model. Anthropic has indicated that this AI presents significant risks to digital security. Earlier this month, after a code leak involving Claude, the company published a statement noting that such AI systems now exceed most human experts in identifying and taking advantage of weaknesses in software. It highlighted potential serious consequences for economic stability, public well-being, and defense. The gathering occurred as bank leaders were already in the capital for an industry association event, with invitations extended to executives from banks deemed critical to the financial system—those whose failure or interruption could threaten overall stability. Attendees included Goldman Sachs CEO David Solomon, Bank of America’s Brian Moynihan, Citigroup’s Jane Fraser, Morgan Stanley’s Ted Pick, and Wells Fargo’s Charlie Scharf, as reported by Bloomberg, which broke the story. JPMorgan Chase’s Jamie Dimon received an invitation but could not participate. In his yearly message to investors released this week, Dimon stated that digital security continues to be a primary concern and that artificial intelligence is likely to heighten these dangers. Anthropic reports that its unreleased Mythos model has identified thousands of flaws in various programs and apps, leading the firm to restrict access to a select group of companies, such as Amazon, Apple, and Microsoft. This is the first instance where Anthropic has limited distribution of one of its offerings. Networking firms Cisco and Broadcom, along with the Linux Foundation—which supports the open-source Linux operating system—have also been granted access. The development raises fears that malicious actors might employ similar technologies to guess passwords or breach data protection measures. Anthropic noted that some vulnerabilities detected by Mythos date back 27 years and appear to have gone undetected by developers or security teams until now. The discussion follows the US government’s recent classification of Anthropic as a risk in supply chains, a designation the company is challenging legally. The Federal Reserve, Anthropic, the involved banks, and the Treasury did not provide statements in response to inquiries from Bloomberg.
