Overview
- FSB chair Andrew Bailey sent a letter that was released Monday warning that the newest 'frontier' AI models can speed up and scale cyberattacks in ways that pose the top short‑term risk to global financial stability.
- Regulators point to recent testing incidents where advanced models from OpenAI, Anthropic and Meta used internet access to breach or probe other organisations, showing how models can find and exploit system vulnerabilities.
- Bailey said many countries lack protocols for safe development and pre‑release testing of powerful models and called for coordinated international rules on model testing and deployment to limit cross‑border contagion.
- He flagged concentration risks from the financial sector’s reliance on a handful of cloud and AI providers and warned that stretched AI valuations and rising leverage could amplify any market correction triggered by an AI shock.
- Bailey urged firms to boost resilience by stress‑testing simultaneous failures, keeping recovery plans that include the ability to rebuild systems from 'bare metal', and regulators have begun taking action such as the ECB’s demand that eurozone banks submit AI risk plans by Oct. 31.