UK financial regulators are holding urgent talks with the Bank of England, the Financial Conduct Authority, HM Treasury, and the National Cyber Security Centre to brief major British banks, insurers, and exchanges on the cyber risks posed by Anthropic's latest AI model, according to the Financial Times. The meeting is expected to take place within the next two weeks, Reuters reported, citing two people briefed on the discussions.
The move follows an extraordinary few days in Washington. On Tuesday, U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened CEOs from Citigroup, Morgan Stanley, Bank of America, Wells Fargo, and Goldman Sachs at the Treasury to warn them about the same model. JPMorgan CEO Jamie Dimon was unable to attend. Kevin Hassett, the White House national economic adviser, confirmed the meeting on Fox News, saying Bessent and Powell "went through the cyber risks to make sure that they were aware of them," Reuters reported.
The model in question is Claude Mythos Preview, a frontier AI system Anthropic unveiled on April 7th alongside Project Glasswing, an industry coalition that includes Amazon, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, NVIDIA, Palo Alto Networks, and the Linux Foundation. Anthropic committed up to $100 million in usage credits for the model and $4 million in direct donations to open-source security organizations.
The reason for the urgency is in the technical report. Anthropic's red team found that Mythos Preview can identify and exploit zero-day vulnerabilities in every major operating system and web browser, according to the company's technical blog. The vulnerabilities it has found are often decades old and survived extensive human review and automated testing. In one benchmark where the previous generation Opus 4.6 achieved near-zero success at autonomous exploit development, Mythos Preview generated working exploits 181 times out of several hundred attempts and achieved register control in 29 more cases. In internal tests across roughly 7,000 entry points into open-source repositories, Mythos Preview achieved full control flow hijack on ten separate, fully patched targets — something neither Sonnet 4.6 nor Opus 4.6 managed.
Anthropic did not explicitly train Mythos Preview for these capabilities. They emerged as a downstream consequence of general improvements in code, reasoning, and autonomy. "The same improvements that make the model substantially more effective at patching vulnerabilities also make it substantially more effective at exploiting them," the company wrote.
The company has restricted broad access. Roughly 40 organizations, including Microsoft and Google, have been granted access through Project Glasswing to use the model for defensive security work. Anthropic said it is sharing findings so the whole industry can benefit. The company added that over 99% of the vulnerabilities found have not yet been patched, and it would be irresponsible to disclose details before fixes are available, following a coordinated vulnerability disclosure process.
The UK meeting is the first formal multi-agency response by financial regulators to a frontier AI model. Canada held a separate meeting on Friday involving the Finance Ministry, the Bank of Canada, and bank executives, where Anthropic's model was also a topic of discussion. Reuters could not immediately verify the UK report.
Anthropic declined to comment for this article. The Bank of England also declined to comment. The Treasury, NCSC, and FCA were not immediately available for comment.
What makes this episode unusual is the source. Anthropic is the lab most associated with cautious, safety-first behavior in AI development. That such a lab is now at the center of an unprecedented multi-government response to a model's capabilities suggests the industry has crossed a threshold that no longer depends on which lab is building the models. The capabilities are real and they are proliferating. The question regulators and industry are now racing to answer is whether defenders can move faster than everyone else.
