Anthropic’s growth draws scrutiny
U.S. financial regulators summoned bank chiefs to warn them about cyber risks tied to Anthropic’s latest AI model, treating the technology as an infrastructure-level concern rather than a simple vendor tool. (theguardian.com) At the same time, reporting says Anthropic’s commercial scale is surging — run-rate revenues reportedly rose toward $30bn and the company is exploring custom AI chips to control cost and capacity. (thenextweb.com)
Washington did not call in bank chiefs over a chatbot subscription. On April 10, the New York Times reported that Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell gathered big-bank executives to discuss cyber risks tied to Anthropic’s new Claude Mythos Preview model. (nytimes.com) Anthropic itself says Claude Mythos Preview is its “most capable frontier model” and says it is not making the system generally available. The company says the model will instead be used with a limited set of partners in a defensive cybersecurity program. (anthropic.com) The reason banks got pulled into this is simple: software bugs are like unlocked side doors in a skyscraper, and a model that finds them faster can help defenders or attackers. Anthropic says Mythos showed a “striking leap” over its previous top model on capability tests, including cybersecurity evaluations. (anthropic.com) Anthropic launched that limited program on April 8 under the name Project Glasswing. The company says the group includes Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. (anthropic.com) That partner list explains why regulators are treating this less like office software and more like infrastructure. JPMorganChase is in the project, and Anthropic says the goal is to secure “the world’s most critical software,” which puts banks, cloud platforms, and core business systems in the same frame. (anthropic.com) At the same time, Anthropic is getting big enough that its supply chain matters on its own. Reuters reported on April 9 that the company’s annualized revenue run rate had climbed past $30 billion, up from about $9 billion at the end of 2025. (usnews.com) When an artificial intelligence company grows that fast, chips stop being a back-office purchase and start looking like fuel. Reuters reported that Anthropic is now exploring whether to design its own chips, though the effort is still early and the company has not committed to a design or team. (usnews.com) Right now Anthropic already depends on other companies’ silicon. Reuters said it uses Google’s Tensor Processing Units, which are specialized chips built for artificial intelligence workloads, as well as Amazon chips to train and run Claude. (usnews.com) Anthropic also just locked in more future capacity instead of waiting for the market to loosen up. On April 6, the company said it signed a new agreement with Google and Broadcom for multiple gigawatts of next-generation Tensor Processing Unit capacity starting in 2027. (anthropic.com) So the same week produced two signals that usually arrive years apart. Regulators treated Anthropic’s newest model like a possible financial-stability issue, while Anthropic treated its own computing needs like those of a utility planning power plants years in advance. (nytimes.com) (anthropic.com) Anthropic’s own safety policy helps explain the split-screen. In February 2026, the company published Version 3.0 of its Responsible Scaling Policy, calling it a voluntary framework for mitigating catastrophic risks from increasingly capable systems, and Mythos is the clearest sign yet that those risks are moving from theory into bank-board conversations. (anthropic.com)
