BoE's Bailey demanded Anthropic brief the FSB on Mythos after warning the AI model "may have found a way to crack the whole cyber-risk world open." The Claude chatbot maker will present its unreleased Mythos Preview cybersecurity model to the Financial Stability Board, the G20 body that coordinates global financial rules, following a direct request from Bailey, who chairs the FSB. The Financial Times reported the development Monday, citing people familiar with the plan.
Bailey flagged his concerns at Columbia University in April, days after Anthropic announced Mythos. "You wake up to find that Anthropic may have found a way to crack the whole cyber-risk world open," Bailey said.
"The issue is: to what extent is this new version of the product going to be able to identify vulnerabilities in other systems which can be exploited for cyberattack purposes."
Mythos is a specialized AI system designed to detect decades-old vulnerabilities in web browsers, infrastructure and software. It has not been publicly released and is currently under controlled testing access. The Pentagon confirmed it is deploying the model under a restricted program to find and patch vulnerabilities across US government systems.
One incident during testing escalated those concerns. The AI system reportedly escaped a controlled digital environment and directly contacted an Anthropic employee to disclose software vulnerabilities, overriding restrictions placed on the model, according to Storyboard18.
Regulators worry Mythos could expose weaknesses in bank cyber defenses faster than institutions can patch them, a critical risk for a banking industry reliant on legacy technology. In the US, Treasury Secretary Scott Bessent and Federal Reserve Chair Jay Powell reportedly met major American banks to discuss emerging AI-linked cyber threats. In the UK, AI minister Kanishka Narayan acknowledged governments should be concerned about capabilities demonstrated by models such as Mythos.
Anthropic has restricted access to around 40 organizations, primarily in the US. Companies granted access include Amazon, Microsoft and JPMorgan Chase. The company agreed not to expand broader distribution after concerns raised by the White House. The briefing comes as AI-enabled cyber attacks increased 89% in 2025 compared with the prior year, according to CrowdStrike. The International Monetary Fund has warned that advanced AI systems could elevate cyber threats into a macro-financial risk affecting global economic stability.
South Africa, an FSB member, is represented on the body by Reserve Bank governor Lesetjsa Kganyago and national treasury director-general Duncan Pieterse.
