Anthropic Model Sparks Alarm

Washington moved faster than Wall Street this week. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell called bank chief executives into an urgent meeting after Anthropic released a new model called Mythos and warned it could raise cyber risk across finance. (reuters.com) (usnews.com) The meeting happened in Washington on Tuesday, April 7, while many large-bank chiefs were already in town for other events. Bloomberg reported that the chief executives of Citigroup, Morgan Stanley, Bank of America, Wells Fargo, and Goldman Sachs attended, while JPMorgan Chase chief executive Jamie Dimon did not. (cnbc.com) (usnews.com) Anthropic’s own warning is what made this different. On April 7, the company said Mythos Preview was not being released broadly because it had become unusually strong at finding and exploiting software flaws, which is the digital equivalent of spotting a weak lock and then picking it. (anthropic.com 1) (anthropic.com 2) Anthropic said the model had already found thousands of high-severity vulnerabilities, including some in every major operating system and every major web browser. Those are the basic layers that run laptops, servers, phones, and the browser tabs where bank staff and customers do daily work. (anthropic.com) (usnews.com) That is why banks care even though Anthropic is not a bank vendor. A model that can help defenders patch code can also help attackers chain together bugs faster, and large banks sit on payment systems, trading pipes, customer data, and identity systems that break in expensive ways when one weak point is missed. (anthropic.com) (nextgov.com) Anthropic tried to get ahead of that by limiting access. The company said more than 40 organizations that build or maintain critical software infrastructure would get Mythos for defensive work, and its launch partners included Amazon Web Services, Apple, Cisco, CrowdStrike, Google, JPMorgan Chase, Microsoft, Nvidia, and Palo Alto Networks. (anthropic.com 1) (anthropic.com 2) It also put real money behind that plan. Anthropic said it would provide up to $100 million in usage credits and $4 million in donations to open-source security groups through a program called Project Glasswing. (anthropic.com 1) (anthropic.com 2) Investors heard a second message: if one model is already this good at coding and vulnerability hunting, more software work may get automated faster than expected. Reuters reported that the Standard and Poor’s 500 Software and Services Index fell 2.6% on Thursday and is down 25.5% so far in 2026. (reuters.com) (usnews.com) Some of the sharpest drops hit security names that normally benefit when cyber risk rises. Reuters said Cloudflare, Okta, CrowdStrike, and SentinelOne fell between 4.7% and 7.7% in morning trading, while Zscaler dropped 8.6%. (reuters.com) (finance.yahoo.com) The strange part of this story is that the alarm did not come from a hack or a breach. It came from a model maker saying, in effect, that the next generation of artificial intelligence is good enough at breaking software that the safest first move is a controlled release and a call to the people who run the financial system. (anthropic.com) (bloomberglaw.com)