Financial regulators race to build their own AI tools

TL;DR:

  • Banks and regulators must move quickly to adopt AI as the technology supercharges cybersecurity risks, Swiss regulator FINMA’s president has warned.
  • An international forum covering 95% of global markets is jointly building supervisory tools, with a recent hackathon focused on crypto oversight.
  • Experience with models such as Anthropic’s Mythos has exposed fresh operational risks for financial institutions.

Financial watchdogs are scrambling to arm themselves with AI to keep pace with the firms they supervise. Marlene Amstad, president of Swiss regulator FINMA and chair of an international supervisory-technology forum, told Reuters that banks and regulators alike must adopt new tools rapidly as AI accelerates cybersecurity threats. “As hackers move faster, banks must adapt by patching vulnerabilities more rapidly,” she said.

Supervisors build for the AI era

FINMA helped create a forum within the International Organization of Securities Commissions — a standard-setter whose members cover around 95% of global markets — to promote regulators’ adoption of AI. Around 100 policy and technology specialists met this week for a hackathon aimed at building shared tools for crypto-market supervision, with regulators exploring whether safeguards could be embedded directly into digital-asset systems. Amstad said experience with models such as Anthropic’s Mythos had exposed vulnerabilities and revealed new AI-related operational risks, after the US ordered the model’s export suspended on security grounds. China’s 360 Security Technology, meanwhile, said it has built a domestic answer to Mythos.

The UK’s own regulator is moving in step. The Financial Conduct Authority has been rethinking its role for the AI age, exploring agentic AI as a “first responder” to scan roughly a billion rows of market data a day and trialling AI on live financial-crime intelligence with Palantir. The common thread is regulators recognising they cannot oversee AI-powered markets with pre-AI tooling.

Looking forward

For UK financial firms, the shift carries a double edge. Better-equipped supervisors should mean faster detection of genuine risk, but also sharper scrutiny of how firms deploy AI themselves. As watchdogs from Zurich to London build their own capabilities, the gap between regulated and regulator is narrowing — and “Switzerland must retain access to the most advanced AI models”, Amstad noted, a reminder that supervisors face the same frontier-access constraints as the firms they police.

Share this article