TL;DR
The Prudential Regulation Authority, the Financial Conduct Authority and the Bank of England published a joint set of effective practices on AI and cyber risk on 19 November. Their central message is that AI has cut the cost, skill and time attackers need to find and use software flaws, so regulated firms, including financial market infrastructure, must find, rank and fix weaknesses far faster than before. The practices are not formal rules.
Why the regulators acted
The paper argues that capable AI can read source code and map routes into a system quickly enough for criminals to exploit a weakness within hours or days, where it once took weeks or months. It cites the UK AI Security Institute, which has found that in some tested areas AI performance doubles every eight months, and that open models have closed much of the distance to closed ones over two years. Humans still drive the most advanced attacks, the regulators add.
Their answer is a shift away from periodic patching cycles. Instead of finding flaws every so often and working through them, firms are told to adopt “exposure management”: continuously tracking every misconfiguration, identity, cloud service and third-party link an attacker could use, including AI agents and open-weight models a firm depends on.
What firms are asked to do
Several themes stand out:
- Board speed. Cyber risk is described as a core business risk. Boards should check not only that controls exist but that they will work during an incident, and may need to accept that fixing a critical flaw quickly can disrupt a service.
- Harness engineering. Having a strong AI model does not by itself improve defence. What matters, the paper says, is the “harness” of tools, workflows and checks around it, which may decide how many flaws a firm actually finds and fixes.
- Human oversight. Reviewers need the authority and time to override or stop AI, and firms should test for “automation bias”, where staff accept outputs uncritically. Kill switches, already used in algorithmic trading, are cited as a model.
- Shared defence. The regulators warn that AI-related cyber risk could threaten financial stability through suppliers and software that many firms share, and point to the Cross-Market Operational Resilience Group as a route for collective response.
Looking forward
The publication sits alongside the regulators’ existing supervisory statement on cyber resilience rather than replacing it, and is meant to be applied in proportion to a firm’s size and risk. It follows remarks by the PRA’s Katharine Braddick on banks trimming AI exposure. In our view, the stress on decision speed in the boardroom is the part most firms will find hardest, because, as the regulators put it, the nature and speed of board reporting and decision-making need to evolve to match an automated attacker.