Bank of England Governor Andrew Bailey, who chairs the global financial-risk watchdog, has invited the company to present to G20 finance ministries and central banks. The model that worried him is the model now being explained.
Anthropic is preparing to brief the Financial Stability Board on the cybersecurity vulnerabilities its Mythos model has been identifying in the global financial system, the Financial Times reported on Monday, citing people familiar with the plan.
The briefing has been requested by Andrew Bailey, the Bank of England governor and FSB chair, and will be delivered to G20 finance ministries and central banks under the board’s umbrella.
Bailey is the right person to have made the call. In an April 15 speech at Columbia University, he named Mythos by name, alongside the Gulf escalation, as one of the two events that had moved cyber up regulators’ risk ranking ‘faster than any other category in recent years’.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
‘It would be reasonable to think that the events in the Gulf are the most recent challenge to us in this world,’ Bailey said, ‘until, I think it was last Friday, you wake up to find that Anthropic may have found a way to crack the whole cyber risk world open.’
Mythos was announced last month, but has not been released. Anthropic describes it as a cybersecurity model designed to surface long-standing vulnerabilities in browsers, infrastructure, and software, and has claimed it has already found thousands of high-severity flaws across major operating systems and browsers.
When directed to develop working exploits against those flaws in internal testing, the model reportedly succeeded on the first attempt in more than 83% of cases. The dual-use implications were obvious to regulators before the model went on the road.
The FSB briefing closes a regulatory arc that began with Bailey’s speech and has since pulled in jurisdictions on three continents.
UK banks were given their own Mythos briefing within days of the Columbia remarks. The Federal Reserve and US Treasury convened the chief executives of major American banks on the same risk shortly afterwards.
Australia’s securities regulator joined the watch-list in early May. Euro-area finance ministers raised access demands of their own, and Mythos has since been delivered into Japanese megabanks, as we reported last week.
What the briefing will not resolve is the access question. Mythos is currently being made available under ‘Project Glasswing’, the controlled-access programme Anthropic has set up to limit who can run the model and against what.
Roughly 40 to 50 organisations have early access, including AWS, Apple, Google, Microsoft, Nvidia, Cisco, and JPMorgan. Bank supervisors outside that list have, on the public record, been pressing for either direct access or a regulator-mediated equivalent, and the FSB session will be the first time those requests are coordinated rather than nationally.
The political subtext is harder than the technical one. The model being briefed to G20 financial regulators is a US-headquartered AI system whose distribution and military access have, separately, been the subject of an ongoing fight between Anthropic and the Trump administration.
Regulators on the receiving end will be aware that the company appearing in front of them has, in parallel, been negotiating its export profile with Washington. Anthropic and the FSB had not responded to Reuters requests for comment by Monday afternoon, and the timing of the actual briefing has not been publicly disclosed.
Mythos is the first publicly disclosed AI system that has, on its developer’s own account, found exploitable vulnerabilities in ‘every major operating system and web browser’.
The FSB briefing is, in essence, the first time the global financial-supervision community will sit in a room together and ask what the practical implications of that finding are.
The question Bailey raised at Columbia, of how much harder the model has made the attack side relative to the defence side, is still the central one.
