Finance ministers from the world's leading economies gathered for emergency discussions this week as Mythos AI — Anthropic's latest foundation model — demonstrated an unprecedented capacity to discover and exploit security weaknesses across major operating systems and financial networks. The crisis response at the International Monetary Fund signals the first time AI capabilities have prompted coordinated government action at this high level.
Canadian Finance Minister François-Philippe Champagne told the BBC that Mythos dominated talks among G7 finance ministers at the IMF meeting. "Certainly it is serious enough to warrant the attention of all the finance ministers," he said. "The difference is that the Strait of Hormuz - we know where it is and we know how large it is... the issue that we're facing with Anthropic is that it's the unknown, unknown."
Anthropic reports that Mythos has already identified security vulnerabilities in key operating systems, financial infrastructure, and major web browsers — prompting the company to offer banks and governments advance access before any public release.
"We have to understand it better, and we have to understand the vulnerabilities that are being exposed and fix them quickly," Barclays CEO CS Venkatakrishnan told the BBC, characterizing the development as emblematic of "what the new world is going to be" — an increasingly interconnected financial system with both expanded possibilities and novel attack vectors.
The UK's AI Security Institute published the only independent assessment of Mythos Preview, noting its ability to "exploit systems with weak security posture" while suggesting it may not represent a dramatic improvement over Claude's predecessor, Opus 4. However, the report acknowledged that "more models with these capabilities will be developed."
Bank of England governor Andrew Bailey emphasized the systemic risk: "The consequence could be that there is a development of AI, of modelling, which makes it easier to detect existing vulnerabilities in sort of core IT systems, and then obviously cyber criminals - the bad actors - could seek to exploit them."
The US Treasury has already begun coordinating with major banks to encourage preemptive system testing before Mythos reaches public availability. Financial industry sources indicate another prominent US AI company may soon release a similarly powerful model without equivalent safeguards — accelerating the timeline for institutional response.
James Wise, chair of the Sovereign AI unit backed by £500m in government funding, framed Mythos as "the first of what will be many more powerful models" capable of exposing system vulnerabilities. His unit is "investing in British AI companies that are tackling that - companies working in AI security and safety."
"We hope the models that expose vulnerabilities are also the models which will fix them," Wise told BBC's Today Programme — articulating the dual-use challenge that has finance ministers scrambling for regulatory frameworks that don't yet exist.
The emergency nature of this week's discussions reflects a broader institutional void: no existing oversight body possesses clear authority over foundation models with these capabilities, leaving finance ministers to coordinate improvised responses to what Champagne termed "the unknown, unknown."
