Anthropic to brief financial‑stability authorities on Claude Mythos' cyber‑relevant flaws

Anthropic is preparing to brief the Financial Stability Board on cyber vulnerabilities identified by its Claude Mythos model, according to reports published on May 18. The company has declined to release Mythos publicly because of concerns that the model’s capabilities could be misused by hackers, the Guardian reported. Reuters separately reported that the briefing will focus on vulnerabilities in the global financial system exposed by the model. The move puts a global finance watchdog at the center of a debate that had largely been framed as an AI-safety and product-release question. Business Insider reported on May 18 that Anthropic and OpenAI have helped trigger a wider scramble among security teams as more advanced coding and cyber-capable AI systems spread through industry. The Financial Stability Board, or FSB, is the G20-backed body that coordinates work on risks to the global financial system. Its 2026 work programme says it plans an October report on “sound practices for AI adoption, use, and innovation.” ### Why is the Financial Stability Board involved at all? The FSB’s remit is systemic risk, not consumer technology. Its 2026 work programme lists vulnerabilities assessments, digital innovation, artificial intelligence and operational resilience among this year’s priorities. Andrew Bailey, the Bank of England governor, is serving as FSB chair, according to Bank of England materials on the FSB’s March 2026 payments summit. (msn.com) The IMF sharpened the backdrop on May 7, when Tobias Adrian, Tamas Gaidosch and Rangachary Ravikumar wrote that AI-enabled cyberattacks could create “correlated failures” affecting payments, confidence and financial intermediation. The same IMF post cited Anthropic’s controlled release of Mythos as an example of how quickly cyber risks are increasing. (fsb.org) ### What seems to have changed in Anthropic’s handling of Mythos? Reuters reported on May 18 that Anthropic is set to discuss cyber vulnerabilities in the global financial system identified by Mythos with FSB members. The Guardian reported that Anthropic chose not to release the model publicly because of fears it could be abused by hackers. Those reports indicate the company is favoring targeted disclosure to institutions over a broad public launch. (imf.org) Reuters also reported on May 18 that Anthropic revised an earlier position and would allow users of its Mythos cybersecurity model to share information about cyber threats with others exposed to similar vulnerabilities. That suggests Anthropic is widening the circle for defensive coordination while still limiting general access to the model itself. (msn.com) ### What does Mythos reportedly do that has regulators paying attention? Reuters reported on May 12 that U.S. banks were rushing to fix scores of IT weaknesses flagged by Mythos. CNBC reported on May 8 that banks, software companies and governments had been jolted by claims that the model found thousands of previously unknown vulnerabilities in software infrastructure. (msn.com) The IMF said on May 7 that advanced AI tools can reduce the time and cost needed to identify and exploit vulnerabilities, increasing the odds that attackers could discover and target weaknesses in widely used systems at the same time. That is the mechanism by which a model issue becomes a financial-stability issue: common software, common dependencies and simultaneous exposure across institutions. (msn.com) ### Why does this matter for banks and public-sector buyers? The immediate issue for banks is not whether they can buy Mythos. It is whether procurement, access controls and disclosure expectations around cyber-capable AI tools are tightening. Reuters’ reporting says the briefing concerns vulnerabilities in the global financial system, which puts regulated institutions directly in scope. (imf.org) The FSB’s published 2026 agenda gives authorities a place to fold those findings into broader work on AI, resilience and vulnerabilities. The next visible checkpoint is the board’s planned October 2026 report on sound practices for AI adoption, use and innovation. (fsb.org) (msn.com)