What the UK regulators published
On 19 November 2026, the Bank of England, the Prudential Regulation Authority (PRA) and the Financial Conduct Authority (FCA) jointly published effective practices for managing cyber risk relating to artificial intelligence. They released a separate set on quantum computing at the same time, alongside a formal consultation on cyber resilience guidance. The regulators say the AI practices are non-binding and sit outside their formal expectations. Firms are meant to apply them in proportion to their size, how much they rely on AI for important services, and how much autonomy and access their AI systems have.
The consultation paper (CP17/26 and GC26/9) closes on 26 February 2027, and the final guidance is due next year. The AI and quantum practices aren't open for formal consultation. The Bank of England's announcement says the regulators want industry feedback on whether this kind of practical material is useful.
The figures behind the publication explain why regulators are paying attention. Looking two years ahead, 88% of those who answered a joint Bank of England and FCA survey thought the cyber abilities of frontier AI would change their risk profile to a high or moderate degree, while 64% picked patching and vulnerability management as the area where frontier AI had lifted management's focus the most.
What the 13 practices ask for
The document's central argument is that AI has cut the cost, skill and time needed to find and exploit software flaws. The regulators write that exploitation can now happen in "hours or days rather than weeks or months". Finding vulnerabilities is no longer the slow part of an attack. The pressure now falls on how quickly a defender can prioritise, validate, fix and recover.
Several of the 13 practices will be familiar to any IT team. Others are new because they treat AI systems as part of the attack surface:
- Asset maps should include AI systems, APIs, cloud services and identities, with AI agents listed among those identities. The regulators suggest building on existing AI and model inventories, outsourcing registers and software bills of materials.
- Vulnerability ranking should weigh active exploitation, impact on important business services, third-party reliance and end-of-life status. Slow remediation and growing backlogs count as operational risks in their own right.
- Where no patch exists, firms should use temporary controls such as traffic filtering or virtual patching, and log every unfixed issue with a named owner and a dated plan.
- Supplier contracts should cover AI disclosure, patching service levels and notice when a vendor changes the AI inside its product.
- Incident playbooks should be written for AI incidents and set out escalation, decision authority and who can shut a system down. Recovery plans should cover compromised models, agent states and embeddings.
- Testing should include hallucination and prompt injection scenarios, and should time how fast leaders reach decisions as well as how fast engineers patch.
The practices also ask for human oversight that scales with risk, kill switches and logs for AI systems, and secure development checks that scan AI-generated code. None of these items is exotic. Taken together, the list amounts to a well-run vulnerability and change process with AI systems added to its scope.
Why a UK document matters in New Zealand
New Zealand regulators are on the same track, with less detail published so far. On 15 October 2026, the Reserve Bank and the Financial Markets Authority set out supervisory expectations on cyber risk linked to frontier AI. The two regulators named three risks: higher cyber risk, governance and oversight challenges, and growing third-party and concentration risk. They expect boards and senior managers to know where AI is used in the business and who is accountable for its risks. The release gives no patching timelines or supplier requirements, and the Reserve Bank and FMA haven't published a practice document comparable to the UK one.
For government agencies, the National Cyber Security Centre published cyber readiness advice for the frontier AI era in June 2026, adapted from an Australian advisory. The NCSC advice points agencies back to existing controls: the NZISM, the Minimum Cyber Security Standards and confirmed executive accountability. Like the Reserve Bank and FMA release, the NCSC page sets no patch timeframes and no AI inventory requirement.
The UK practices fill a gap that NZ guidance leaves open for now. A New Zealand bank, insurer or fund manager that wants to show the Reserve Bank or FMA it has acted on the October expectations can use the UK list as a credible, regulator-written benchmark. Firms outside financial services can use it too. Many NZ businesses rely on the same cloud platforms, managed service providers and software vendors as UK banks, and an attacker using AI to find flaws in a shared product won't distinguish between a London bank and a Hamilton manufacturer.
Where NZ firms are likely to fall short
From the systems we see in New Zealand and Australian organisations, three of the practices will stretch most mid-sized firms. AI agents rarely appear in identity inventories, because staff create them inside Microsoft 365, Salesforce or ERP add-ons, and nobody registers the service accounts the agents use. Supplier contracts signed before 2024 seldom say anything about vendors adding AI features or about patch service levels. Patch backlogs tend to be tracked as an IT hygiene metric, and the person accountable for the business service the unpatched system supports often never sees them.
The UK document's point about leadership decision speed deserves attention here. Many incident response plans assume a chief executive or board chair can be reached within a day. If exploitation now takes hours, a plan with a 24-hour approval step for taking a customer-facing system offline is already too slow.
Practical next steps
- Add AI to your asset and identity registers. List every AI agent, copilot and model integration, with the account each one runs under and the data and tools it can reach.
- Set patch timeframes by exposure. Agree maximum fix times for internet-facing and actively exploited flaws, and report the backlog to the owner of each business service as well as to IT. Our ITIL 4 service management work covers how change and vulnerability processes fit together.
- Record unfixed issues formally. Every vulnerability without a patch needs a compensating control, a named owner and a review date.
- Review supplier contracts at renewal. Ask for AI disclosure, notice of AI changes and patching service levels, including from your managed service provider.
- Write an AI incident playbook. Name who can switch off an agent or AI feature, how to do it, and what manual fallback replaces it.
- Test decision speed. Run a tabletop exercise in which leaders must decide to isolate a key system within two hours, and note where approvals stall.
- For Reserve Bank and FMA regulated entities, map your current controls against the UK list and keep the result as evidence of your response to the October expectations. For help setting up AI governance and accountability, see our AI consultancy services.
The UK practices aren't binding anywhere, including in the UK. They're still the most concrete regulator-written checklist on AI-driven cyber risk published so far, and NZ firms can use them now without waiting for local guidance to catch up.
Sources
- Effective practices: cyber risk and resilience management relating to artificial intelligence (Bank of England / PRA)
- Cyber resilience guidance consultation and AI and quantum effective practices (Bank of England)
- UK financial watchdogs clarify cyber expectations as attacks rise (FCA)
- RBNZ and FMA outline supervisory expectations on cyber and AI risk (FMA)
- Cyber readiness in the frontier AI era (NCSC New Zealand)
- Microsoft and organisations across industries launch AI Transparency Initiative (Microsoft)
Want to know what this means for your organisation? Our team in Auckland can show you your top three opportunities in a free 30 minute assessment.
Get my free assessment How we help