Anthropic: cyber alarm and chips
U.S. financial and regulatory leaders met urgently with Wall Street over concerns that Anthropic’s latest AI could raise cyber risks for banks, signalling heightened scrutiny of model deployment in critical sectors. At the same time, reports say Anthropic is exploring custom AI chips as compute and cost needs grow, a sign that model firms are treating silicon strategy as a core option rather than an exotic side project. Those two threads together show Anthropic is at the intersection of infrastructure scale and regulatory attention. (x.com, thenextweb.com, news.abplive.com)
On April 7, Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell called Wall Street leaders to Washington over one company’s new artificial intelligence model: Anthropic’s Mythos. Bloomberg reported that officials were worried the model could sharply raise cyber risk if tools that find software flaws also become tools that exploit them. (bloomberg.com) Anthropic says Mythos is so strong at finding vulnerabilities in software and computer systems that it is not releasing the model broadly. Bloomberg reported that banks were being urged to test it internally, with JPMorgan Chase named and Goldman Sachs, Citigroup, and Bank of America also getting access or expecting access. (bloomberg.com) That is a strange position for regulators to be in. The same model can work like a building inspector that finds weak locks before a break-in, or like a burglar carrying a master key, depending on who has it and what safeguards are wrapped around it. (bloomberg.com) The concern was big enough that it spread beyond the United States within days. Bloomberg reported that the Bank of Canada and major Canadian lenders held a similar meeting on April 10 after the Washington discussion earlier that week. (bloomberg.com) At the same time, Anthropic is dealing with a second problem that sounds less dramatic but may be just as important: chips. Reuters reported on April 9 that Anthropic is exploring whether to design its own artificial intelligence chips because the supply of chips used to train and run advanced models remains tight. (usnews.com) A custom chip is the engine under the hood of an artificial intelligence system. If you build your own engine, you can tune it for your own model, lower long-term costs, and depend a little less on the few suppliers that control the fastest hardware. (cnbc.com) Anthropic is not starting from scratch on infrastructure. Amazon has invested a total of $8 billion in the company and made Amazon Web Services its primary cloud and training partner, while Anthropic said on April 6 that it had also expanded a compute partnership with Google and Broadcom. (anthropic.com, anthropic.com) The scale explains why chips are now a boardroom issue instead of an engineering footnote. Anthropic said its revenue run rate has passed $30 billion, up from about $9 billion at the end of 2025, and that more than 1,000 business customers are now each spending over $1 million a year on an annualized basis. (anthropic.com) Anthropic has also become too large to be treated like a niche lab. The company said in February that it raised $30 billion in Series G funding at a $380 billion post-money valuation, which means decisions about one model release or one hardware bet now ripple into banks, cloud providers, and governments. (anthropic.com) Put those two threads together and the picture gets clearer. Anthropic is being pulled from both ends at once: officials want tighter control over what its most capable models can do, and the company wants tighter control over the chips those models need to exist at all. (bloomberg.com, usnews.com)
