Treasury and Fed held emergency talks on Anthropic risk
A Bloomberg-cited report says the U.S. Treasury secretary and the Fed chair convened urgent meetings with Wall Street about cyber risks tied to Anthropic’s newest AI model. The briefing suggests regulators and financial firms are treating model-related cyber exposures as immediate systemic concerns rather than a distant policy debate. (x.com)
Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell called major Wall Street chiefs into an urgent meeting in Washington this week over one company’s new artificial intelligence model, according to Bloomberg reporting published on April 9 and updated on April 10. The trigger was Anthropic’s latest system, which officials fear could sharply raise cyber risk for banks. (bloomberg.com) The unusual part is who showed up. The Treasury Department handles financial stability and sanctions, while the Federal Reserve oversees big banks and payment plumbing, so a joint warning from Bessent and Powell signals they see this as a system problem, not a niche tech story. (bloomberg.com) Anthropic unveiled Claude Mythos Preview on April 7 and said the model is especially strong at finding weaknesses in software. Anthropic also said it was limiting access because the same skill that helps defenders patch bugs could help attackers find doors that nobody knew were unlocked. (red.anthropic.com) In plain English, this kind of model works like a tireless code auditor that can read huge amounts of software and point to the exact line where a break-in might start. If that auditor lands in the wrong hands, it can become a map for burglars instead of a checklist for guards. (red.anthropic.com) Anthropic said Mythos had already identified thousands of severe software flaws across major systems through a program it calls Project Glasswing. The company said it was keeping the model to a small group of organizations rather than releasing it broadly. (red.anthropic.com) Banks are a special case because they run on old and new software at the same time. A large bank can have decades-old core systems, modern mobile apps, outside vendors, and nonstop links to payment networks, which means one newly exposed weakness can spread trouble far beyond one company. (federalreserve.gov) That is why regulators care about “systemic” cyber risk. In finance, systemic risk means a failure at one important firm or one shared network can spill into funding markets, payment systems, and customer access across the country. (treasury.gov) The banking alarm did not appear out of nowhere this week. Bloomberg reported in late March that cybersecurity stocks fell after concern spread that an Anthropic model being tested might help hackers get around current defenses. (bloomberg.com) Anthropic has been warning about this tradeoff itself. In its April 7 write-up, the company said Mythos shows unusually strong cyber capability, and CNBC reported the same day that Anthropic limited the rollout over fears the model could be used for cyberattacks. (red.anthropic.com) (cnbc.com) Put together, the message from Washington is that artificial intelligence risk has moved from chatbots saying odd things to machines that can speed up the hunt for real software breakpoints. When the Treasury secretary and the Federal Reserve chair are briefing bank chiefs before any public disaster, they are treating that shift as a live-fire financial stability issue now, not a theory for later. (bloomberg.com)
