Anthropic faces scrutiny and chip rumors

Anthropic is suddenly dealing with two very different problems at once: getting enough chips to keep growing, and convincing Washington that one of its newest models will not supercharge cyberattacks. Those two fights landed in public within the same week of April 2026. (cnbc.com 1) (cnbc.com 2) On the hardware side, Reuters reported on April 9 that Anthropic is exploring whether to design its own artificial intelligence chips, though the effort is still early enough that the company could decide not to build them at all. The immediate reason is simple: advanced artificial intelligence systems need huge volumes of scarce chips to train models and answer user requests. (usnews.com) (cnbc.com) Anthropic is not starting from zero on hardware, because it already spreads its workloads across Amazon Web Services Trainium chips, Google Tensor Processing Units, and Nvidia graphics processors. The company said on April 6 that this mix gives it “better performance and greater resilience” than relying on one supplier. (anthropic.com 1) (anthropic.com 2) That diversification matters because Anthropic is tied closely to Amazon and Google at the same time. Amazon said in November 2024 that Anthropic had named Amazon Web Services its primary training partner, while Anthropic said this week that Google Cloud and Broadcom are expanding capacity for Claude inside the United States. (aboutamazon.com) (anthropic.com) The scale behind that chip question is enormous now. Anthropic said on February 12, 2026 that it raised $30 billion in Series G funding at a $380 billion post-money valuation, and said the money would go toward research, products, and infrastructure. (anthropic.com) Then the other thread hit. Anthropic announced Claude Mythos Preview on April 7 through its security unit, saying the model is unusually strong at finding severe software bugs and can reverse-engineer exploits on closed-source software. (anthropic.com) (axios.com) Anthropic did not launch Mythos like a normal chatbot. It limited access to a small group of cybersecurity and technology partners, and Axios reported that the company was withholding a public release because it feared the model could be misused to find and exploit security flaws. (anthropic.com) (axios.com) By April 9, concern had moved from the tech industry to the top of the U.S. financial system. Multiple reports said Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell met major bank chiefs to discuss cyber risks tied to Mythos and the possibility that stronger bug-finding tools could make digital break-ins easier. (straitstimes.com) (livemint.com) Anthropic has spent years selling itself as the safety-minded alternative in artificial intelligence, so the contrast is sharp here. The same company that is building a giant domestic compute footprint is also testing a model it says can uncover high-severity vulnerabilities faster than existing systems. (anthropic.com 1) (anthropic.com 2) Put together, the chip story and the Mythos story describe the same phase of Anthropic’s life. It is no longer just a lab renting compute and releasing models; it is becoming infrastructure big enough to consider custom silicon, and powerful enough that Treasury and the Federal Reserve are discussing its products with Wall Street in real time. (cnbc.com) (straitstimes.com)