UK Regulators Issue AI Cyber-Risk Practices as Financial Institutions Face Machine-Speed Threats

The Bank of England, Prudential Regulation Authority (PRA) and Financial Conduct Authority (FCA) have jointly published new effective practices for managing cyber risk and resilience relating to artificial intelligence (AI). Published on 19 November 2026, the document addresses how AI is changing the cyber-threat landscape while also providing financial institutions with new tools to strengthen their defences. The practices are non-binding and complement, rather than replace, existing regulatory expectations on cyber risk and resilience.

A central concern is the speed at which AI can enable cyberattacks. The regulators note that advanced AI can analyse source code, identify vulnerabilities and discover attack paths at speeds that could allow threat actors to exploit weaknesses within hours or days rather than weeks or months. This is pushing firms away from traditional periodic vulnerability management towards continuous exposure management, where firms continuously identify, prioritise, remediate and recover from exploitable weaknesses.

The framework recommends that firms map exposures across AI systems, APIs, cloud services, external connectors, identities, hardware and software, third-party dependencies and open-source components. Firms are also encouraged to use AI and automation for vulnerability identification, triage and remediation, including automated patching where feasible. However, the regulators stress that AI should augment rather than replace human judgement, particularly for important decisions and high-risk activities.

One notable concept introduced in the document is “harness engineering”—the tools, workflows, controls, data and operating environments surrounding an AI system. A well-designed harness should determine what information the AI receives, what tasks it can perform, how outputs are validated and how findings reach cybersecurity and engineering teams. The regulators emphasise that simply obtaining access to advanced AI does not automatically create effective cyber defence; firms need appropriate workflows, controls, validation and human review around the technology.

The guidance also places significant emphasis on identity and access management, secure software development, third-party risk and incident recovery. Firms should consider phishing-resistant multi-factor authentication, least-privilege access, zero-trust principles, segmentation, continuous logging and AI-specific threat modelling. Incident-response plans may need to address prompt injection, model poisoning, jailbreaking, malicious inputs and attacks involving highly autonomous AI agents. Recovery procedures may also need to account for compromised models, agent states, memory and AI embeddings.

For boards and senior management, the message is particularly important: AI-related cyber risk is becoming a core business and resilience issue rather than merely a technology problem. The regulators say leadership needs assurance not only that cyber controls are designed appropriately, but that they will actually work during an incident. This comes as cyber incidents reported to the FCA in October 2026 were 52% above the previous 18-month peak recorded in May 2025, reinforcing the urgency of strengthening cyber resilience across financial services.

Want to deepen your expertise beyond today’s news?

#Riskmanagementnews

author avatar
RMA INDIA

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.