Anthropic to Brief Global Financial Regulators on Cyber Risks from Unreleased Mythos AI

A Model Considered Too Dangerous to Release

Anthropic is preparing one of the most unusual briefings in the recent history of financial regulation. The company will appear before the Financial Stability Board to walk global regulators through cybersecurity weaknesses in the worldwide financial system uncovered by its unreleased Mythos AI model. The session was requested directly by Bank of England Governor Andrew Bailey, who chairs the FSB, and the findings will be circulated to G20 finance ministries and central banks.

Mythos, announced in April as Claude Mythos Preview, is a specialised system built to hunt long-standing vulnerabilities in software, browsers, and critical infrastructure. Anthropic has said it can identify exploitable flaws across every major operating system and web browser, and reports that it produced working exploits on the first attempt in more than 83% of internal tests. Rather than releasing it, the company has limited access to roughly 40 trusted organisations, including Amazon, Apple, Google, and JPMorgan, through a coordinated disclosure programme called Project Glasswing.

Regulators Sound the Alarm

The FSB briefing follows a joint statement on 14 May from HM Treasury, the Bank of England, and the Financial Conduct Authority warning that the cyber capabilities of frontier AI now exceed what a skilled human practitioner could achieve, at greater speed, larger scale, and lower cost. The regulators urged firms and financial market infrastructures to strengthen defences, sharpen vulnerability management, and consider deploying AI-enabled protections of their own.

In the United States, Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Bessent convened an urgent meeting with major bank chief executives in April to discuss the cyber risks linked to Mythos. European lawmakers have separately pressed the European Commission to act, and Japan is negotiating direct access to the model to bolster its national defences.

Systemic Risk Enters the Conversation

The planned briefing marks a turning point. Frontier AI capabilities are no longer being framed purely as a technology policy issue. They are being treated as a potential source of systemic financial risk on par with liquidity shocks and contagion. Bringing a private model maker before the FSB underscores a growing consensus among global financial authorities that systems capable of both finding and weaponising software flaws require coordinated oversight at the highest levels. The episode also raises a thorny new question for the industry: when a model is too dangerous to ship, who decides who gets to use it, and on what terms?