Urgent Anthropic cyber meeting

Washington pulled top bank chiefs into a closed-door meeting this week because regulators think the next cyber shock may not start with a human hacker at all. Bloomberg and CBS reported that Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell called in Wall Street leaders at Treasury’s headquarters in Washington on Tuesday, April 7, to discuss risks tied to Anthropic’s new Mythos model. (bloomberg.com) (cbsnews.com) The reason banks care is simple: modern finance runs on software the way cities run on electricity. Bloomberg reported that every bank invited was classified as systemically important, which means a software failure or breach at one of them can spill into the rest of the financial system. (bloomberg.com) Anthropic says Mythos is unusually good at finding a software vulnerability, which is a hidden mistake in code that can act like an unlocked window in a building. In its April 7 technical write-up, Anthropic said Mythos could identify and then exploit zero-day vulnerabilities, meaning flaws unknown to the vendor, in every major operating system and every major web browser during testing. (red.anthropic.com) Anthropic also said many of the flaws it found were old and hard to spot, including one now-patched OpenBSD bug that had survived for 27 years. That detail matters because it suggests the model is not just scanning for obvious errors but digging up problems that human teams missed for decades. (red.anthropic.com) That is why Anthropic did not put Mythos on the open market. On April 7, the company launched Project Glasswing and said only a small set of partners would get early access, including Amazon Web Services, Apple, Cisco, Google, Microsoft, NVIDIA, Palo Alto Networks, and JPMorganChase. (anthropic.com) Anthropic said it also extended access to more than 40 additional organizations tied to critical software infrastructure, while committing up to $100 million in usage credits and $4 million in donations to open-source security groups. The pitch is to hand the new lock-picking machine to defenders first so they can reinforce doors before criminals get the same tool. (anthropic.com) The company is also trying to slow down how details of newly found flaws spread. Anthropic’s March 6 disclosure policy says it aims to notify vendors quickly, follow a 90-day disclosure clock in normal cases, and move faster on actively exploited critical bugs with a 7-day patch or mitigation target. (anthropic.com) The second part of the story is happening on developers’ laptops, not in Washington conference rooms. Claude Code, Anthropic’s coding agent, can read a codebase, edit files, run tests, and use command-line tools, which makes it powerful for software teams and also sensitive from a security standpoint. (anthropic.com) At the same time regulators were worrying about stronger cloud models, developers were sharing ways to run Claude Code against local models on their own machines. Ollama said in January 2026 that its software had added Anthropic Messages application programming interface compatibility, making it possible to connect tools like Claude Code to local or self-hosted models instead of Anthropic’s servers. (ollama.com) (docs.ollama.com) That creates two opposite pressures at once. Governments and banks are treating frontier coding models like a possible systemic cyber risk, while developers are pushing the same style of tools toward private, offline, harder-to-monitor use on personal hardware. (bloomberg.com) (docs.ollama.com) Treasury told CBS that more coordination meetings are planned, which means officials do not see this as a one-day panic. The immediate question is no longer whether artificial intelligence can write code faster than humans, but whether it can now find the weak seams in the software that banks, browsers, and operating systems have been trusting for years. (cbsnews.com)