Anthropic is providing the Financial Stability Board with an update on the findings from Mythos.

      Bank of England Governor Andrew Bailey, who leads the international financial-risk oversight body, has invited the company to present to the G20 finance ministries and central banks. The model that caused him concern is the one currently being discussed. Anthropic is gearing up to inform the Financial Stability Board (FSB) about the cybersecurity vulnerabilities that its Mythos model has been identifying within the global financial landscape, according to a report by the Financial Times on Monday, citing sources familiar with the matter.

      The presentation has been requested by Andrew Bailey, the FSB chair and Bank of England governor, and will be conducted for G20 finance ministries and central banks as part of the board’s activities. Bailey is well-positioned to initiate this action. In a speech given on April 15 at Columbia University, he specifically mentioned Mythos, alongside the situation in the Gulf, as two events that have elevated cybersecurity in regulators’ risk assessments ‘more rapidly than any other category in recent years’.

      “It would be reasonable to believe that the Gulf events are the latest challenge we face,” Bailey remarked, “until, I think it was last Friday, you wake up to discover that Anthropic might have discovered a way to disrupt the entire cybersecurity landscape.” Mythos was unveiled last month but has yet to be released. Anthropic characterizes it as a cybersecurity model aimed at revealing long-standing vulnerabilities in browsers, infrastructure, and software, boasting that it has already identified thousands of high-severity flaws across major operating systems and web browsers.

      In internal testing designed to develop successful exploits against these flaws, the model reportedly achieved success on the first attempt over 83% of the time. The dual-use implications were apparent to regulators even before the model was publicly presented. The FSB briefing marks the conclusion of a regulatory journey that began with Bailey’s speech and has since involved jurisdictions across three continents.

      UK banks received their own briefing on Mythos shortly after Bailey's Columbia remarks. The Federal Reserve and US Treasury subsequently convened the CEOs of major American banks to discuss the same risks. Australia’s securities regulator joined the watch-list in early May. Finance ministers in the Euro area also raised their own concerns, and Mythos has since been provided to major Japanese banks, as previously reported.

      However, the briefing will not address the matter of access. Mythos is currently accessible through ‘Project Glasswing’, a controlled-access initiative established by Anthropic to limit who can operate the model and under what conditions. Around 40 to 50 organizations have early access, including AWS, Apple, Google, Microsoft, Nvidia, Cisco, and JPMorgan. Regulators from outside this group have been publicly advocating for either direct access or a regulator-mediated equivalent. The FSB session will mark the first time these requests are consolidated rather than addressed individually by each nation.

      The political implications are more complex than the technical ones. The model being presented to G20 financial regulators is a US-based AI system, whose distribution and military access have been subjects of ongoing contention between Anthropic and the Trump administration. Regulators are likely conscious that the company appearing before them has simultaneously been negotiating its export profile with Washington. By Monday afternoon, neither Anthropic nor the FSB had responded to Reuters' requests for comments, and the exact timing of the briefing has not been publicly revealed.

      Mythos is the first publicly acknowledged AI system that, according to its developer, has discovered exploitable vulnerabilities across ‘every major operating system and web browser’. The FSB briefing is essentially the first occasion for the global financial supervision community to convene and discuss the practical implications of this finding. The critical question raised by Bailey at Columbia—how much more difficult the model has made the offense compared to defense—remains a central issue.