Full Breakdown
Anthropic’s Claude Mythos AI Model Briefed to Global Financial Regulators
5/19/2026, 4:48:27 AM
Briefing to the Financial Stability Board
On 18 May 2026 Anthropic briefed the Financial Stability Board (FSB) on its Claude Mythos AI model at the request of Bank of England Governor Andrew Bailey, who chairs the FSB. The model remains unreleased, with access limited to a group of partners.
Stakeholders and Partnerships
Stakeholders include Anthropic; the FSB; the Bank of England, the Financial Conduct Authority and HM Treasury; the International Monetary Fund; and partners in Project Glasswing—Apple, Amazon, Microsoft, Nvidia and JP Morgan. The U.S. Department of Defense is deploying Mythos to patch software.
Capabilities and Early Findings
Mythos can locate vulnerabilities in browsers and operating systems. Anthropic reports thousands of severe flaws and a demonstration of exploits across platforms. The UK AI Security Institute recorded a 3-in-10 success on the “cooling-tower” test—first for any model—and noted that cyber-task length has doubled within months. Partners may share findings under responsible-disclosure norms.
Official Statements & Responses
The FSB spokesperson said it “welcomes engagement with Anthropic and other firms on frontier risks to financial stability.” The Bank of England, FCA and HM Treasury warned that frontier AI already exceeds human capabilities, operating faster, at scale and lower cost, and urged stronger cyber hygiene. The IMF cautioned that cyber risk ignores borders, calling for oversight. BoE Governor Andrew Bailey warned that Mythos could expose previously unknown cyber vulnerabilities, likening the discovery to a sudden opening of the cyber-risk world.
Criticism & Opposition
Industry experts argue that most data breaches still arise from weak authentication and unpatched known flaws, indicating that Mythos represents an evolution rather than a revolution in cyber threats. They caution that the model’s capabilities may amplify existing vulnerabilities but do not constitute a new attack vector.
Conflicting Reports & Gaps
Anthropic and some regulators describe Mythos as capable of “supercharging” attacks, while other experts argue its impact is limited by persistent legacy vulnerabilities. The model’s full capabilities remain undisclosed and no public benchmark data exist, leaving a verification gap.
Verbatim Quotes
- “welcomes engagement with Anthropic and other firms on emerging and frontier risks to global financial stability.” — FSB spokesperson
- “Cyber risk does not respect borders. As AI capabilities spread across countries, inconsistent oversight could weaken a globally interconnected system,” — International Monetary Fund
- “Rathi added: “Anthropic has acted fairly responsibly in engaging with authorities” internationally on the risks of its AI.” — Nikhil Rathi, Chief Executive, FCA
- “We fully support our partners sharing findings with each other and companies outside of Glasswing to triage vulnerabilities,” — Anthropic spokesperson
What’s Next
The FSB will issue a report on AI-adoption practices by early June 2026. The Bank of England, FCA and HM Treasury will monitor frontier AI via CMORG and update cyber-resilience guidance. Anthropic’s partners will expand information sharing under the revised Project Glasswing, with further briefings to non-U.S. finance ministries planned.
