Banks warned on Anthropic risk
U.S. officials convened major bank executives to warn about cyber‑risks posed by Anthropic’s newest AI model, treating frontier‑model capabilities as an operational threat rather than abstract policy fodder. The meeting was led by the Treasury secretary and the Federal Reserve chair and was reported as a direct outreach to large American banks. (nytimes.com / theguardian.com)
Scott Bessent and Jerome Powell quietly pulled top Wall Street chiefs into the Treasury Department on Tuesday, April 7, for a warning that was not about interest rates or bank runs. It was about a new Anthropic model called Mythos and the chance that it could make cyberattacks faster and cheaper. (bloomberg.com) The people in the room reportedly included chiefs from Bank of America, Citigroup, Wells Fargo, Goldman Sachs, and Morgan Stanley, while JPMorgan Chase chief executive Jamie Dimon was invited but did not attend. The meeting was described as a direct warning to large banks to harden their systems against a new kind of machine-helped intrusion. (kval.com) (cnbc.com) Anthropic had just announced Mythos Preview on April 7, but it did not release the model to the public. The company said only a small group of partners would get access because the system was unusually good at finding software flaws that criminals could also use. (anthropic.com) (techcrunch.com) A software flaw is like a hidden side door in a building that even the owner does not know exists. A “zero-day” flaw is worse, because the vendor has had zero days to fix it once someone discovers it. (bloomberg.com) Anthropic said Mythos found thousands of zero-day vulnerabilities in testing, including flaws in every major operating system and every major web browser. The company also said the model could identify and then exploit those flaws when a user asked it to. (anthropic.com) (cbsnews.com) That is why banks got the call first. Big banks run on old and new software at the same time, move trillions of dollars through connected systems every day, and sit on customer data that makes them prime targets for ransomware crews and state-backed hackers. (reuters.com) (americanbanker.com) Anthropic says Mythos is being shared instead through a defensive program called Project Glasswing. The initial group includes Microsoft, Amazon, Apple, CrowdStrike, Palo Alto Networks, and about 40 other companies that will test the model for security work rather than general use. (cnbc.com) The unusual part is not just that officials worried about artificial intelligence in general. It is that the Treasury secretary and the chair of the Federal Reserve treated one named model from one named company as a live operational risk to the financial system. (theguardian.com) (nytimes.com) For years, frontier artificial intelligence was mostly discussed in terms of future regulation, copyright fights, and job losses. This week, Washington handled it more like a hurricane warning for bank networks: private briefings, named targets, and instructions to prepare before the damage starts. (nytimes.com) (bloomberg.com)
