NICE's AI Delivery Plan for the NHS: What Compliance Teams Need to Know

NICE published a delivery plan in August 2026 setting out how it intends to use artificial intelligence across its own work, and how it will evaluate the growing wave of AI-enabled health technologies coming through the system. For anyone in healthcare compliance or clinical governance, this matters because it signals exactly what evidence NICE will expect before an AI tool gets anywhere near routine NHS use.

What the plan actually covers

NICE's delivery plan has three distinct strands. First, using AI in evidence generation and evaluation — applying AI tools to NICE's own process of reviewing clinical evidence. Second, evaluating health technologies that themselves incorporate AI, which is the strand most relevant to providers and manufacturers bringing AI-enabled products into the NHS, and which echoes the evidence-led approach already reshaping NHS patient safety governance more broadly. Third, exploring AI to improve NICE's own internal productivity, aiming to speed up how quickly guidance gets published.

The scope of AI technologies NICE is explicitly considering is broad: diagnostic imaging and fracture-detection tools, predictive risk-identification models, clinical decision-support software, administrative AI that summarises information and streamlines processes, and digital tools for managing long-term conditions. This isn't a narrow policy aimed only at diagnostic AI — it spans clinical and administrative applications alike.

What NICE will actually require as evidence

The plan sets out safeguards that AI-enabled products will need to demonstrate: improved patient outcomes, value for money, consistent performance across different populations, safety and trustworthiness, and evidence that's transparent, reproducible and explainable. That last point is worth dwelling on — "explainable" evidence is a materially higher bar than simply showing a tool works, since it requires being able to account for why a model reaches the conclusions it does, not just that its outputs correlate with good outcomes in a trial.

NICE has also confirmed it will coordinate with the MHRA on regulation, meaning AI-enabled health technologies will need to satisfy both NICE's evidence and value standards and the MHRA's separate medical-device regulatory requirements — two distinct approval tracks that providers and manufacturers need to plan for in parallel, not sequentially.

Timeline

NICE has committed to publishing an AI best practice methods framework for evidence generation by March 2027, which will presumably give far more procedural detail than the delivery plan itself currently offers. Separately, NICE has set an internal target of halving the staff time needed to produce guidance by 2030, using AI to speed up its own evidence-review processes — an ambitious efficiency goal that, if achieved, could meaningfully shorten how long it takes new treatments and technologies to reach NICE-approved guidance generally.

Why compliance and clinical governance teams should care now

It's tempting to file this under "future regulatory framework, check back in 2027." But two things make earlier engagement worthwhile. First, any organisation already piloting or evaluating an AI tool — a triage system, a decision-support tool, an administrative AI assistant — should start collecting the kind of outcome, safety and explainability evidence NICE has flagged now, rather than retrofitting evidence collection once the methods framework lands and inspection or procurement processes start referencing it. Second, procurement teams evaluating AI vendors can already use NICE's stated criteria — outcomes, value, consistency across populations, safety, explainability — as a practical due-diligence checklist today, well ahead of any formal framework requiring it.

This sits alongside the broader shift toward outcomes-based, evidence-led regulation showing up across UK and Irish healthcare standards more generally — see our guide to the CQC Single Assessment Framework for a related example of how UK inspection regimes are formalising evidence expectations elsewhere in the system.

How this compares with AI governance in other Learnsignal-covered sectors

Finance and accounting have been through a broadly similar reckoning with AI governance over the past couple of years, and the parallels are instructive. Auditors and finance teams adopting AI tools have had to grapple with the same core questions NICE is now formalising for healthcare: can the tool's outputs be explained and defended, does it perform consistently across different scenarios rather than just the ones it was trained on, and is there a clear chain of accountability when something goes wrong. Healthcare compliance teams building AI governance frameworks now don't need to invent this from scratch — the underlying principles of explainability, consistent performance and clear accountability are increasingly common across every sector adopting AI in a regulated environment, even though the specific regulators and evidence requirements differ. Building a governance framework flexible enough to satisfy NICE's criteria today, and whatever the March 2027 methods framework adds, is a more durable approach than treating each new AI tool's approval as a one-off exercise.

FAQs

Does this mean AI tools can't be used in the NHS yet? No — AI tools are already used across the NHS. The delivery plan sets out how NICE will formalise its evaluation approach going forward, not a moratorium on current use.

Is this only about diagnostic AI? No — the plan explicitly covers diagnostic, predictive, clinical decision-support and administrative AI applications, plus digital tools for long-term condition management.

When will the detailed evidence framework be published? NICE has committed to publishing its AI best practice methods framework for evidence generation by March 2027.

Does NICE approval replace MHRA regulation for AI medical devices? No — NICE has confirmed it will coordinate with the MHRA, meaning AI-enabled health technologies need to satisfy both bodies' separate requirements.

Learnsignal's Healthcare Compliance & CPD courses cover the evidence and governance standards healthcare organisations are expected to meet as clinical technology continues to evolve.