Urgent Anthropic risk meeting
U.S. financial leaders held an emergency meeting to discuss cyber and financial risks from Anthropic’s newest AI model, signalling regulators and markets are treating advanced AI as an operational threat. Treasury Secretary Scott Bessent and Fed Chair Jerome Powell joined Wall Street figures to assess the issue, underscoring rapid official attention to model safety in finance. (x.com)
On Tuesday, Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell called bank chief executives into an urgent meeting after officials concluded Anthropic’s newest model could sharply raise cyber risk for the financial system. Reuters and Bloomberg both reported the meeting on April 9, 2026, and tied it to fears about Anthropic’s latest release. (reuters.com, bloomberglaw.com) The reason this landed in Washington instead of a normal bank security office is simple: banks run on old, sprawling software, and a tool that finds hidden flaws faster than humans can turns one weak server into a systemwide problem. The Federal Reserve sits at the center of U.S. financial plumbing, so Jerome Powell showing up meant this was being treated as an operational threat, not a science demo. (finance.yahoo.com, federalreserve.gov) Anthropic’s model is called Claude Mythos Preview, and the company did not put it on the open market. Anthropic said on April 7 that it would keep the model inside a restricted program called Project Glasswing instead of making it generally available. (anthropic.com, red.anthropic.com) Anthropic says the model can identify and exploit zero-day vulnerabilities, which are software bugs defenders have not patched because they do not know about them yet. In its own write-up, the company said Mythos Preview found vulnerabilities across every major operating system and every major web browser during testing. (red.anthropic.com, anthropic.com) That is why Anthropic paired the model with a closed defense coalition instead of a public launch. Project Glasswing started with partners including Amazon Web Services, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, NVIDIA, Palo Alto Networks, and the Linux Foundation. (anthropic.com, cnbc.com) Anthropic also put real money behind the lockdown. The company said it is committing up to $100 million in usage credits for Mythos Preview and another $4 million in direct donations to open-source security groups working on critical software. (anthropic.com) Banks were always going to be near the front of the line because they are giant targets with giant attack surfaces. JPMorganChase is already a launch partner in Glasswing, so the same firm was sitting inside Anthropic’s defense project while Washington was simultaneously warning Wall Street that the model’s offensive potential could outrun today’s safeguards. (anthropic.com, reuters.com) Anthropic’s own safety documents say Mythos Preview is its most capable frontier model to date and marks a sharp jump from Claude Opus 4.6 on many benchmarks. When the company that built the system says it is too dangerous for broad release, regulators do not need to imagine the downside from scratch. (anthropic.com, red.anthropic.com) This is the part that changed fast: three days after Anthropic unveiled Mythos Preview on April 7, the story was no longer about a product launch. By April 10, it had become a Treasury-and-Federal-Reserve problem, which is about as close as artificial intelligence gets to being treated like a live financial stability issue. (anthropic.com, bloomberg.com)
