Anthropic has agreed to brief global finance ministries and central banks on cyber vulnerabilities identified by its latest AI model, Mythos, following a request by Andrew Bailey as concerns grow over threats to the international banking system.

The Financial Times reported on Monday that Bailey, who chairs the Financial Stability Board, asked Anthropic to discuss the capabilities of its Claude Mythos Preview model with the watchdog’s members. The FSB brings together finance ministries, central banks and regulators from G20 economies including the US, UK, China, Japan and Germany.

According to the Financial Times, regulators are increasingly concerned that advanced AI systems could expose weaknesses in banks’ cyber defences and increase the risk of large-scale attacks on financial infrastructure. Anthropic said last month that Mythos had “found thousands of high-severity vulnerabilities, including some in every major operating system and web browser”, warning that the consequences for economies and national security “could be severe”.

Anthropic has restricted access to Mythos because of concerns about how the technology could be misused. Around 40 organisations, including Amazon, Microsoft and JPMorgan Chase, have been granted access to help identify and patch software vulnerabilities, while the company has agreed not to distribute the model more broadly following a request from the White House.

Cyber security specialists have warned the system could enable more sophisticated attacks against banks that still rely on ageing technology infrastructure. Bailey raised similar concerns during an appearance at Columbia University in April, saying: “Anthropic may have found a way to crack the whole cyber risk world open.”

Bailey added that the key issue was “to what extent is this new version of the product going to be able to, in a sense, identify vulnerabilities in other systems which can be exploited for cyber attack purposes”.

The FSB is preparing a report on “sound practices” for AI adoption in financial services, which is expected to be released for consultation next month. Both the FSB and Anthropic declined to comment on the recent communications.

The International Monetary Fund warned earlier this month that emerging AI models could “elevate cyber risk to a potential macro-financial shock”. IMF officials wrote in a blog post that “cyber risk does not respect borders” and said developing economies could face greater exposure because of weaker defences and limited resources.

---