The National Commission into the Regulation of AI in Healthcare has now closed its open call for evidence with more than 770 responses, a third of them from patients and members of the public. The headline finding from Professor Henrietta Hughes, who chairs the commission’s Health Systems Working Group, is that trust is the primary barrier to AI adoption in UK healthcare. For NHS trust executives planning their next AI deployment, that single finding reframes the problem. The constraint on adoption is no longer the technology, the integration work, or even the budget. It is whether patients and clinicians will let the system be used at all.
TL;DR
- The commission’s evidence base shows broad public support for AI as clinician support, but cautious resistance to AI making high-stakes decisions independently.
- Trust, not capability, is the binding constraint — and the public is asking for reform of regulation, not its overhaul.
- Two specific concerns dominate: post-market surveillance after device approval, and legal accountability when AI is involved in clinical decisions.
- NHS trust deployment plans that focus on tooling without explicit answers to those two questions will stall, regardless of clinical efficacy.
- The summer recommendations will set a baseline; trusts that build trust-mechanics into procurement now will move faster when they land.
What 770 submissions actually tell us
The composition of the response matters as much as the volume. Roughly 257 of the 770 submissions came from patients and members of the public — a level of unsolicited public engagement that is unusual for a regulatory call for evidence. The MHRA layered on sector roundtables involving 117 clinicians and 30 organisations, and partnered with National Voices and the Health Foundation to reach groups typically absent from these processes, including young people and those with learning disabilities.
This is not a technical consultation that happened to receive public comment. It is a public consultation that happened to receive technical comment. The implication for NHS trusts: the political ground for AI deployment is being shaped by patient voice, not by clinician voice or vendor voice. Operational plans drafted on the assumption that the binding stakeholder is the clinical director or the IT director are working from the wrong stakeholder map.
Strategic Reality: The public is not opposed to NHS AI. It is opposed to NHS AI that operates without clear human oversight, transparent accountability, and post-deployment monitoring. The deployment question is no longer “will patients accept this?” but “what does the deployment have to look like for patients to accept it?”
The substantive finding is even more useful. The public is broadly supportive of AI for two specific use cases: earlier diagnosis and the reduction of administrative burden on clinicians. Both have direct line-of-sight to felt patient experience — shorter waits, more clinician time, faster answers. The public is markedly less supportive of AI in high-stakes autonomous decision-making. The line is not “AI yes” or “AI no”. The line is between AI that augments a clinician and AI that replaces a clinician’s judgement.
Why trust is the binding constraint
In a privately-funded healthcare system, capability is often the binding constraint on adoption: if the technology works, customers buy it. In a publicly-funded system that operates on consent — clinical consent from staff, social consent from patients, political consent from Parliament — trust is the binding constraint. A clinically excellent AI system that loses social or clinical consent does not get deployed at scale, no matter how good its evidence base.
| Constraint | Where it usually binds | Where it binds in NHS AI |
|---|---|---|
| Technical capability | Early adoption phase | Largely solved for diagnostic and admin use cases |
| Procurement budget | Commercial healthcare | Real but secondary; trusts have AI funding routes |
| Clinical evidence | Regulatory approval | Necessary but insufficient |
| Post-market surveillance | Long after deployment | Now a pre-deployment question |
| Legal liability | Disputed only after harm | Now a board-level deployment gate |
| Public and clinician trust | Late-stage rollout | The first gate, not the last |
The post-market surveillance and liability concerns named in the commission’s findings collapse the timeline. Concerns that traditionally became live only after harm occurred are now being asked at the procurement stage. NHS trusts that cannot answer them in their business cases will find that their AI programmes stall not at the technical pilot, but at the governance and information-governance committees that sign off deployment.
Critical Context: Two specific gaps surface repeatedly in the commission’s evidence: ongoing safety monitoring after a device receives initial approval, and clarity on legal accountability when AI is involved in a clinical decision. Both are operational questions a trust-level playbook can answer; neither is answered by the AI vendor’s marketing material.
What an NHS trust deployment playbook now needs to address
A deployment playbook built only around clinical efficacy, integration architecture, and benefits realisation is no longer sufficient. The commission’s findings imply five additional sections that should sit alongside the technical plan, each with named owners, named artefacts, and named decision-points.
1. Trust positioning. A clear, written position on where AI sits in the clinical pathway: support tool, decision aid, or autonomous decision-maker. The public consensus is that AI is a support mechanism. A trust deploying AI in a way that the public would describe as autonomous decision-making — even if the trust would not — is taking on political risk it has not priced.
2. Post-market surveillance. A live monitoring approach for every deployed AI system that continues after MHRA approval, with clear thresholds for retraining, recalibration, suspension, or withdrawal. Surveillance of model drift, of demographic performance gaps, and of error patterns over time should be a named operational responsibility, not a research project.
3. Liability and accountability. A documented position on who is accountable when AI contributes to a clinical decision that goes wrong. The commission’s evidence repeatedly identifies this as a vital protection for both patients and practitioners. Trusts that defer this question to vendors or to the centre will find their clinicians refusing to use the tool.
4. Patient-facing transparency. A consistent answer to the patient question “was AI involved in my care, and how?” — not buried in a privacy notice, but deliverable verbally by the clinician at the point of care. The public is not asking for AI to be removed from healthcare; it is asking to know when and how it is being used.
5. Underrepresented-group inclusion. Performance monitoring across groups that the commission specifically prioritised through its National Voices and Health Foundation outreach: young people, those with learning disabilities, and other typically underrepresented populations. AI systems trained on overall performance often fail differentially in subgroups. The commission’s outreach has primed expectations on this point.
Implementation Note: Of these five, post-market surveillance is usually the weakest in current deployment plans. It is also the area where public concern is most specific. A trust playbook that names a post-market surveillance owner, a frequency, and a kill-switch criterion will pass scrutiny that competing playbooks fail.
Stakeholder impact: who has to do what differently
Different stakeholders inside an NHS trust face different consequences from the trust-as-binding-constraint reframe. The roles that bore the deployment work in earlier waves are not the roles that will bear it now.
| Stakeholder | What changes | What to do now |
|---|---|---|
| Trust board | AI deployments need explicit consent and risk pricing at board level | Add AI deployments as a standing board agenda item with named risk owner |
| Caldicott Guardian / SIRO | Trust framework now extends from data to model behaviour | Extend information-governance reviews to include post-deployment model monitoring |
| Clinical director | Clinician acceptance now requires liability clarity | Negotiate vendor-provided liability terms before pilot, not after |
| Patient experience lead | Patient-facing AI transparency is now their territory | Develop point-of-care explanation scripts for AI-assisted pathways |
| Procurement lead | Vendor selection criteria need post-market surveillance terms | Include surveillance plans, kill-switch terms, and subgroup performance reporting in tender |
| Communications lead | Trust comms must precede, not follow, deployment | Sequence patient and public communication ahead of go-live, not after |
The most consequential change is the implicit upgrade of the patient experience lead’s role. In most trusts, that role is downstream of deployment decisions; the commission’s findings imply it should be upstream. A communications plan that lands two weeks before go-live cannot reverse a public expectation set during a national consultation.
Reality Check: Within a single trust, AI deployments often sit across a digital programme, an information-governance committee, and a clinical safety lead — three groups with different cadences and different reporting lines. A unified trust position on AI deployment requires somebody senior to own the integration of all three, not a working group that meets monthly.
Strategic recommendations: a sequenced approach for any NHS trust
There is no single right pace for adoption. The right pace depends on where a trust starts and what its current AI-related governance looks like. A pragmatic three-stage frame:
Stage 1: Trusts with no live clinical AI deployment
Priority is foundational positioning. The commission’s summer recommendations will arrive whether or not a trust has prepared. Use the next three to four months to produce a written trust-level position on AI in clinical pathways, a draft post-market surveillance template, and a vendor-engagement checklist that includes liability terms. The cost of being late is not the absence of AI; it is being shaped by national guidance rather than shaping local interpretation of it.
Stage 2: Trusts with departmental or pilot AI deployments
Priority is consolidation. Pilots in radiology, ophthalmology, or administrative triage often pre-date the trust-wide governance frame they now need. Inventory every deployed and piloted AI tool, map each against the five-section playbook above, and identify the gaps. The pilot may have been launched on a clinical-evidence basis without a post-market surveillance plan or a liability position. That is normal; not closing the gap before the next deployment is not.
Stage 3: Trusts with multiple clinical AI deployments at scale
Priority is operational maturity. The trust now needs a clinical AI committee with executive sponsorship, a published register of deployed AI systems, and quarterly reporting to the board on surveillance findings. The commission’s recommendations in summer will likely accelerate sector-wide expectations; trusts with this maturity already in place will treat them as ratification rather than a new burden.
Strategic Insight: The trusts that will look strongest when the commission reports in summer are the ones that have already answered its likely recommendations in their own internal frameworks. The ones that will look weakest are those that wait to be told what to do.
Priority actions across all stages
- Produce a written trust-level position on AI in clinical pathways, agreed at board level, before the commission reports.
- Add post-market surveillance terms to every AI procurement, and refuse to deploy without them.
- Negotiate vendor liability terms that name the trust’s, the vendor’s, and the clinician’s positions explicitly.
- Develop point-of-care patient explanation scripts for every AI-assisted pathway in production.
- Extend information-governance reviews to cover model behaviour, not just data flows.
- Engage the local Healthwatch and patient experience structures before go-live, not after.
- Sequence sector-engagement timing around the commission’s webinar (20 May) and the summer report.
Four hidden challenges
The visible challenges — clinical evidence, integration cost, training — get most of the attention. Four less visible ones may matter more for whether trust-level AI deployment actually progresses.
Challenge 1: The post-market surveillance staffing gap. Surveillance of deployed AI is a continuous activity that requires statistical and clinical-safety capability the typical NHS trust does not have in its current establishment. Vendors often offer monitoring as a service, which is not the same thing — independent surveillance is the patient-protection mechanism, not vendor self-reporting.
Mitigation: Establish a regional or shared-service approach to AI surveillance now. Most trusts cannot reach the staffing threshold alone; ICB-level shared surveillance functions are a feasible response.
Challenge 2: The liability vacuum’s chilling effect on clinicians. Even without a national position, clinicians can refuse to use tools whose liability allocation is unclear — and they should. A trust that does not document a clear local liability position will see its AI utilisation rates fall after initial enthusiasm, regardless of the technical performance of the tools.
Mitigation: Issue a written trust-level liability position even where national guidance is absent. Local clarity is better than national ambiguity for utilisation.
Challenge 3: The transparency–trust paradox. Telling patients more about AI involvement in their care can, in the short term, reduce trust rather than increase it — because patients learn it is being used in places they had not realised. Trusts that disclose only after pressure will pay a higher trust cost than trusts that disclose proactively.
Mitigation: Treat transparency as a one-way ratchet. Disclose proactively and consistently from day one of any deployment; do not retrofit transparency under pressure.
Challenge 4: The summer-recommendations cliff. The commission’s recommendations are expected in summer 2026. Trusts that have made deployment decisions on the assumption of current regulation may find some of those decisions need revisiting against new expectations. The cost of revisiting is highest where commitments to vendors are deepest.
Mitigation: Add a “summer 2026 review point” to every AI deployment decision made between now and then, with explicit board awareness that the regulatory ground is shifting.
Warning ⚠️: The most expensive failure mode for an NHS trust is to deploy AI that performs clinically well but is suspended after a public-facing trust failure — a misdiagnosis story, a transparency complaint, an underrepresented-group performance gap that surfaces in the press. The clinical investment is then sunk, the political capital is spent, and the next deployment in the same trust faces a higher trust bar than the first one did.
What the commission’s framing implies for the rest of the public sector
The commission’s work is healthcare-specific, but the framing — trust as the primary barrier, post-market surveillance as a deployment-stage question, accountability as an operational requirement rather than a regulatory one — is portable. Local authorities deploying AI in children’s services, education AI tools used in schools, AI in benefits processing, AI in policing: all face structurally similar trust constraints, often without the benefit of a national commission to clarify the terms.
For NHS trust leadership specifically, the value of being inside a national process now is significant. The webinar on 20 May, with Professor Hughes, the Chair of the National AI Commission, and the Chief Executive of the MHRA, is a low-cost signal of engagement that will be visible to local stakeholders. Trusts that send their digital, clinical-safety, and patient-experience leads, and turn the session into an internal briefing afterwards, will derive operational value from a session that costs them an hour.
Strategic Reality: The commission’s recommendations will land into a sector where some trusts have already done the playbook work and others have not. National guidance does not equalise the gap — it crystallises it. The next eight weeks are the cheapest moment to close it.
Three success factors and a checklist
Three factors will distinguish NHS trusts that move through the next eighteen months of AI deployment well:
- Position before product — a written, board-level position on the role of AI in clinical pathways that precedes individual procurement decisions, rather than emerging from them.
- Surveillance as default — every deployed AI system has a named owner, a defined monitoring cadence, and a documented kill-switch criterion, not as a research artefact but as standard operating procedure.
- Transparency as one-way ratchet — patient-facing disclosure of AI involvement is proactive, consistent, and increases over time; never retrofitted under pressure.
Next steps checklist
- Draft a trust-level written position on the role of AI in clinical pathways
- Inventory every deployed and piloted AI tool against the five-section playbook
- Add post-market surveillance, liability, and subgroup-performance terms to all AI procurement
- Develop point-of-care patient explanation scripts for AI-assisted pathways
- Engage local Healthwatch and patient experience structures before next go-live
- Brief the trust board on the commission’s webinar (20 May) and summer report
- Establish or join a regional AI surveillance shared-service arrangement
- Review existing vendor contracts for liability and surveillance terms; renegotiate at next break
Take Action: The lowest-cost, highest-value step in the next month is a single-page trust-level position on AI in clinical pathways, signed off at board level. It will not solve the deployment problem on its own, but every other action above is easier to take once it exists, and every other action above is harder to take while it does not.
Source citation
UKAuthority. (2026, April 24). Trust is the primary barrier to AI technology adoption in healthcare says public. https://www.ukauthority.com/articles/trust-is-the-primary-barrier-to-ai-technology-adoption-in-healthcare-says-public
The National Commission into the Regulation of AI in Healthcare is chaired in part by Patient Safety Commissioner Professor Henrietta Hughes, with sector roundtables coordinated by the Medicines and Healthcare products Regulatory Agency (MHRA) and outreach delivered with National Voices and the Health Foundation. Final recommendations are expected in summer 2026, with a public webinar scheduled for Wednesday 20 May.
Analysis and editorial framing by Resultsense. We help UK leaders make sense of AI: see our insights archive for further analysis, or contact us to discuss how this shift applies to your organisation.