Finance ministers and central bankers convened emergency discussions after Anthropic’s unreleased “Mythos” model reportedly surfaced software vulnerabilities that could imperil financial infrastructure. Canadian Finance Minister François-Philippe Champagne said the issue dominated IMF meetings in Washington, while Bank of England Governor Andrew Bailey warned of heightened cybercrime risk if bad actors weaponize the tool. Anthropic has limited access to Mythos under a program dubbed Project Glasswing, giving major tech firms and some financial institutions a head start on patching weaknesses; the UK’s AI Security Institute, however, found the model powerful but not dramatically superior to prior systems. Barclays CEO C.S. Venkatakrishnan said banks must rapidly test and fix exposures, as the U.S. Treasury urged major lenders to harden systems ahead of any broader release. Skeptics note AI developers have previously hyped risks when delaying launches, recalling OpenAI’s staggered GPT-2 rollout.
Related articles:
— Guidance on cyber resilience for financial market infrastructures (CPMI‑IOSCO)
— NIST AI Risk Management Framework
