Fed & Treasury summon Wall Street

On Tuesday, April 7, Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell pulled Wall Street leaders into Treasury headquarters in Washington for an urgent meeting about a new artificial intelligence model from Anthropic called Mythos. Anthropic had launched the model that week but held back a broad public release because of fears it could expose previously unknown cybersecurity vulnerabilities. (semafor.com) (news.bloomberglaw.com) The people in the room were not random bank executives. Bloomberg Law reported that many were already in Washington for a Financial Services Forum gathering, and that group represents the chief executives of the eight United States global systemically important banks, the firms regulators treat as critical to the financial system. (news.bloomberglaw.com) (fsforum.com) (federalreserve.gov) The worry was not that banks had started using Mythos in customer chatbots. The worry was that a stronger model could help attackers find weak points in bank systems the way a master lockpicker spots flaws in a door faster than the owner does. (semafor.com) (news.bloomberglaw.com) That is why Powell’s presence mattered. Bloomberg Law reported that officials saw this as a systemic-risk issue, meaning the failure of one giant bank’s operations or controls could spread through payments, trading, funding, and confidence across the rest of the market. (news.bloomberglaw.com) (federalreserve.gov) Bank regulators already have old rulebooks for model risk and vendor risk, but those rulebooks were written before today’s most powerful generative systems arrived. The Federal Reserve and the Office of the Comptroller of the Currency issued model-risk guidance in 2011, and the Federal Reserve, Federal Deposit Insurance Corporation, and Office of the Comptroller of the Currency updated third-party risk guidance in 2023. (federalreserve.gov 1) (federalreserve.gov 2) (occ.gov) Government reports have been warning for almost two years that artificial intelligence in finance could create exactly this kind of problem. In June 2024, Acting Comptroller Michael Hsu said artificial intelligence could pose systemic risks in banking, and in May 2025 the Government Accountability Office said banks were already using artificial intelligence in trading, lending, and customer service while regulators were still relying mainly on existing laws and exams. (occ.gov) (gao.gov) The Financial Stability Oversight Council made the same point in its 2024 annual report. It said major cyber incidents at large financial institutions could disrupt critical operations more broadly, and it pushed for more information sharing and more work to assess and reduce cyber-related threats. (fraser.stlouisfed.org) What changed this week was speed. Kevin Hassett, the director of the National Economic Council, said on Fox News that Anthropic had agreed to hold back a public release until officials had “figured everything out,” which is a rare sign that Washington treated one company’s model launch less like a product event and more like a financial-stability drill. (news.bloomberglaw.com) The concern is already moving across borders. Bloomberg Law reported that the Bank of Canada met major banks and financial firms on Friday about Mythos-related cyber risks, and that the Bank of England was preparing its own meeting with top bank and insurance executives. (news.bloomberglaw.com) For banks, this pushes artificial intelligence out of the innovation lab and into the same bucket as liquidity, capital, and disaster recovery. A model built by an outside company can now be treated like a critical supplier whose failure, misuse, or compromise has to be tested before it can touch core systems. (federalregister.gov) (gao.gov)