Banks Warned About Anthropic
U.S. regulators reportedly convened major bank executives to outline cyber‑risks tied to Anthropic’s latest AI model, signalling official concern about operational impacts. Coverage says the meeting focused on threat vectors and systemic risk as the model gained traction in financial contexts. Separately, reports note Anthropic is exploring custom AI chips and that the company’s run‑rate revenue has passed $30 billion — moves that suggest both product escalation and vertical integration. (nytimes.com) (thenextweb.com)
The strange part of this story is not that banks are testing artificial intelligence. It is that Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell reportedly pulled top bank chiefs into a warning meeting about one company’s new model. (nytimes.com) (bloomberg.com) The model is Anthropic’s Claude Mythos Preview, which Anthropic rolled out in limited form in early April 2026 instead of releasing it broadly. Anthropic said the system was strong enough at finding and exploiting software weaknesses that only a small group of companies would get access at first. (cnbc.com) (nytimes.com) That is the core fear regulators were discussing. A model that can act like a tireless security tester for defenders can also act like a tireless burglar for attackers if the controls fail. (americanbanker.com) (cnbc.com) Banks are a special case because they run payment rails, trading systems, and customer accounts that other businesses depend on every day. A software break at one giant bank can spread into missed payroll, frozen transfers, and market disruptions far beyond that bank’s own servers. (nytimes.com) (bloomberg.com) Anthropic’s own rollout shows how unusual the model is. About 40 companies including Microsoft, Amazon, Apple, CrowdStrike, and Palo Alto Networks were brought into a defensive security program called Project Glasswing to test the system under tighter controls. (cnbc.com) That limited release did not calm officials. The New York Times reported that the U.S. meeting focused on cyberthreat scenarios and operational fallout, which means the concern was not just stolen data but whether core financial plumbing could keep running under pressure. (nytimes.com) The story widened fast. Bloomberg reported that the Bank of Canada and major Canadian financial firms also met on April 10 to discuss the same Anthropic-related cyber risk, which shows this was being treated as a cross-border financial stability issue, not a single Washington briefing. (bloomberg.com) At the same time, Anthropic is getting much bigger much faster. The company said its annualized revenue run rate has passed $30 billion in April 2026, up from about $9 billion at the end of 2025, and Bloomberg reported more than 1,000 business customers are now spending over $1 million a year. (bloomberg.com) (cnbc.com) That growth helps explain the second half of the news. Reuters reported that Anthropic is exploring its own artificial intelligence chips, even though it already uses hardware tied to Google, Amazon, and Nvidia, because buying compute at this scale starts to look like running an airline that does not own enough planes. (cnbc.com) (thenextweb.com) So the same week produced two signals pointing in opposite directions but from the same cause. Anthropic’s models are becoming powerful enough that governments are treating them like infrastructure risk, while Anthropic itself is becoming large enough to think about owning more of the infrastructure underneath them. (nytimes.com) (thenextweb.com)
