G7 economies plan simulated AI attacks on banks as cyber fears rise

At IMF annual meetings in Bangkok, G7 finance ministers agreed to run frontier-AI tabletop exercises simulating major cyber attacks on the financial sector, Japan’s finance minister Satsuki Katayama confirmed after deputies discussed it. An official said the simulations would be created by different G7 financial authorities, could include commercial banks (none named), and would test governments’ existing cyber emergency manuals; scenarios could include AI-enhanced cyber attacks and AI-related infrastructure as a target, likened to flight-simulator testing for pilots. No timeline or full G7 membership list beyond Japan is given; Netherlands finance minister Eelco Heinen (not G7) is quoted on a shared “black box” and the need for cooperation. Backdrop: South Korea’s president said AI models showed signs of use in bank attacks; the UK FCA warned frontier models find flaws faster than firms can fix them; the IMF Global Financial Stability Report named AI-empowered cyber crime among its three biggest perceived stability threats. IMF data: in 2026, in over 70% of cases exploitation occurred before or on the day of vulnerability disclosure, up 20 percentage points from 2025 and from about 20% in 2018. The FSB discussed AI systemic risk with ministers; Bank of England governor and FSB chair Andrew Bailey said regulators had raised “pace and intensity” and called for more AI model testing and higher safety standards. IMF’s Caio Ferreira cautioned that loss of confidence from service-provider disruption could be systemic even without material losses.