Finance ministers and central bankers are scrambling after Anthropic's latest AI model showed it can find and exploit cybersecurity vulnerabilities in financial systems. Claude Mythos, which Anthropic describes as "strikingly capable at computer security tasks," has triggered crisis meetings at the International Monetary Fund gathering in Washington DC. Canadian Finance Minister François-Philippe Champagne told the BBC it's "serious enough to warrant the attention of all the finance ministers," calling it an "unknown, unknown" compared to threats they can actually see coming.
Anthropic hasn't released Mythos publicly. They're giving controlled access to a handful of tech giants through Project Glasswing, an effort to secure critical software. Amazon Web Services, Microsoft, CrowdStrike, and Nvidia all got access. Barclays CEO CS Venkatakrishnan told the BBC the situation is "serious enough that people have to worry." Bank of England governor Andrew Bailey said they're examining what it means for cyber crime risk. The US Treasury has already pushed major banks to test their systems.
The UK's AI Security Institute, the only independent body to test Mythos, found it powerful but not dramatically better than Anthropic's previous Claude Opus 4. Their report said it "can exploit systems with weak security posture," but they expect many more models with these capabilities to appear.
Some cybersecurity experts question whether the panic is justified given how little independent testing has happened. The restricted release strategy has drawn comparisons to OpenAI's staged GPT-2 rollout in 2019, where fears about misuse ultimately looked overblown.
Financial industry sources told the BBC that another major US AI company could soon release a similarly capable model, potentially without the safeguards Anthropic put in place. James Wise, a partner at Balderton Capital and chair of the UK's Sovereign AI unit, put it plainly: "We hope the models that expose vulnerabilities are also the models which will fix them." That's the bet everyone is making right now.