What does BoE/PRA/FCA effective practices on AI cyber risk (Nov 2026) say about AI in banking?

Published Nov 19, 2026 · Last reviewed Nov 19, 2026

On 19 November 2026 the PRA, FCA and Bank of England jointly published effective practices for managing cyber risk and resilience relating to artificial intelligence, describing a world where AI 'is rapidly creating new cyber risks and amplifying existing ones' by cutting the cost, skills and time attackers need to find and exploit vulnerabilities. The document asks firms to move from periodic vulnerability management to continuous exposure management, to prioritise, validate, remediate and recover 'at machine speed', and to keep human oversight meaningful when AI is used in cyber defence. It is non-binding and sits beside, but outside, the joint consultation CP17/26 / GC26/9 on cyber risk and resilience management, which closes on 26 February 2027.

OFFICIAL TEXT: bankofengland.co.uk ↗ · FINAL · UK (BOE / PRA / FCA)

Checked 2026-11-19 · UK (BoE / PRA / FCA), bankofengland.co.uk
DocumentBoE/PRA/FCA effective practices on AI cyber risk (Nov 2026) — Effective practices: Cyber risk and resilience management relating to artificial intelligence (AI)
Issued byBank of England, Prudential Regulation Authority, and Financial Conduct Authority
TypeGuidance
StatusFinal
PublishedNov 19, 2026
Applies toFirms and financial market infrastructures supervised by the PRA, FCA and Bank of England, including UK banks and building societies; non-binding and meant to be used in proportion to each firm's size, complexity and risk profile
Official sourcebankofengland.co.uk ↗
Use casesCybersecurity · Generative & agentic AI · Third-party & vendor AI · AI governance (general)

What are the key points of BoE/PRA/FCA effective practices on AI cyber risk (Nov 2026)?

  • Published jointly by the PRA, FCA and Bank of England on the same day as consultation CP17/26 / GC26/9; the regulators say the effective practices are complementary to, but separate from, the draft supervisory statement and Final Guidance on cyber risk, and are not part of their formal expectations.
  • Uses the OECD definition of an AI system and avoids the term 'frontier AI', while saying the practices matter most for AI systems whose autonomy and capabilities create high cybersecurity risks.
  • Exposure management: map exposures in AI systems (including open-weight models), APIs and model endpoints, external connectors, AI agents as non-human identities and AI embedded in third-party services, and monitor AI-specific attack surfaces such as embeddings, memory modules and tool chains.
  • Prioritise actively exploited vulnerabilities, assets behind important business services, material third-party services, internet-facing assets and end-of-life assets, and consider the aggregate impact if AI systems chain low-level vulnerabilities; remediation speed and vulnerability backlogs are to be treated as operational risks.
  • 'Harness engineering': the tools, workflows, controls and data around an AI system used for cyber defence should let the firm show how outputs were validated and who is accountable for them.
  • Contracts with material third parties may need to cover disclosure of embedded AI, service levels for vulnerability remediation, and prompt notice of changes to AI systems that affect the provider's cyber risk profile.
  • Response and recovery should cover data or model poisoning, prompt injection, jailbreaking, misuse of retrieval-augmented generation pipelines and attacks by highly autonomous AI agents; recovery may mean rolling back fine-tuned models or agent states.
  • Human oversight should be 'meaningful, risk-based and capable of changing outcomes', with human-in-the-loop approval where errors could be serious or irreversible and kill switches of the kind already used in algorithmic trading.

What did BoE/PRA/FCA effective practices on AI cyber risk (Nov 2026) change for banks?

The May 2026 joint statement by the Bank, FCA and HM Treasury warned that frontier AI models change the cyber threat picture. This publication turns that warning into operational detail on exposure management, AI-assisted defence, third-party contracts, incident response and recovery, without creating new rules. It accompanies the regulators' consultation on a supervisory statement and Final Guidance on cyber risk and resilience management, on which the FCA expects to publish final guidance in the first half of 2027.

Are the UK regulators' AI cyber effective practices binding on banks?

No. The PRA, FCA and Bank of England say the effective practices are complementary to, but separate from, their draft supervisory statement and Final Guidance on cyber risk and resilience management, and are not part of their formal expectations. Firms can adopt them in a way that fits their size, complexity and risk profile, and the regulators welcome feedback on them although they are outside the consultation.

What do UK regulators expect banks to do about AI-enabled cyber attacks?

The 19 November 2026 effective practices describe continuous exposure management, risk-based prioritisation with emergency remediation targets, a governed harness for AI-assisted defence, contract terms covering AI embedded in third-party services, phishing-resistant multi-factor authentication against deepfakes and impersonation, AI-specific incident playbooks, and recovery plans that cover AI agents and compromised models.

When does the related UK cyber consultation close?

CP17/26 / GC26/9, the joint PRA, Bank of England and FCA consultation on cyber risk and resilience management published on 19 November 2026, closes on 26 February 2027. The FCA says it expects to publish final guidance in the first half of 2027.

DateDocumentStatus
Oct 26, 20262026 BoE/FCA AI survey — The Bank of England and FCA's 2026 AI SurveyIn force
Oct 19, 2026FCA: AI in financial services — good practice and areas for improvement — AI in Financial Services: Good Practice and Areas for ImprovementIn force
Sep 30, 2026Bailey: Frontier AI and the Question of Governance (Sep 2026) — Frontier AI and the Question of Governance — Governor Andrew BaileyFinal
Sep 2, 2026FCA multi-firm review: Frontier AI and cyber resilience (Sep 2026) — Frontier AI and Cyber ResilienceIn force
Jul 14, 2026HM Treasury Financial Services AI Adoption Plan (Jul 2026) — Financial Services AI Adoption PlanFinal
May 15, 2026BoE/FCA/HMT joint statement on frontier AI and cyber resilience (May 2026) — The Bank, FCA and HM Treasury joint statement on Frontier AI models and cyber resilienceIn force

Follow every document these regulators publish

when one of these regulators moves, the next morning's brief says so · six sourced stories · 7 am PT · free

plus every regulation, bank and agent page update, the morning after · leave any morning