WalledAI enterprise logo
Back to blog
AI ComplianceOctober 7, 20266 min read

UK Healthcare AI Shifts to Lifecycle Oversight by Spring 2027

The UK government has accepted all 44 recommendations from the National Commission into the Regulation of AI in Healthcare, committing to continuous lifecycle oversight of AI-enabled medical devices instead of relying on one-time approvals.

WA

Walled AI Research Team

On this page›

The UK government has accepted all 44 recommendations from the National Commission into the Regulation of AI in Healthcare, committing the Medicines and Healthcare products Regulatory Agency (MHRA) to move from predominantly point-in-time approvals to continuous, real-world lifecycle oversight of AI-enabled medical devices. The shift, announced on 6 October 2026, means that pre-market clearance will become the start of an assurance programme rather than its conclusion, with a full implementation roadmap promised by Spring 2027 and draft guidance on managing device changes expected by December 2026. MHRA has already opened Phase 3 of its AI Airlock regulatory sandbox to test post-market surveillance models that capture how AI behaves after deployment.

What changed in UK healthcare AI regulation?

On 6 October 2026, the government published its formal response to the National Commission's report, accepting every one of the 44 recommendations in full. The Commission, whose report was published on 10 September 2026, concluded that conventional medical-device regulation was designed for relatively static products, whereas healthcare AI may iterate, perform differently in different environments and depend heavily on data, workflows and users. GOV.UK

The government's response turns that analysis into concrete workstreams. Recommendation 5 commits MHRA to explore a lifecycle-based regulatory framework, generating more evidence throughout the product lifecycle and reducing dependence on point-in-time pre-market assessment. Recommendation 6 calls for Predetermined Change Control Plans (PCCPs), with MHRA committing to publish draft guidance for public comment by December 2026. GOV.UK

Post-market surveillance will become far more intrusive. Recommendation 17 demands tailored post-market surveillance, potentially including post-market studies, ongoing performance reporting and escalation when performance degrades even where no reportable incident has occurred. Recommendation 20 strengthens traceability and version control, including unique device identifiers for software and AI devices and consideration of capturing version information in electronic patient records. Recommendations 18 and 19 propose stronger reporting mechanisms, including improvements to the Yellow Card system and a public-facing database of medical-device adverse incidents. GOV.UK

Parallel to this policy work, MHRA has opened Phase 3 of the AI Airlock, explicitly prioritising post-market surveillance and lifecycle oversight. The sandbox is funded through April 2029, with individual engagements expected to last six to twelve months. Participants must investigate how surveillance can capture deployment-context factors such as local infrastructure, workflow integration, data availability and quality, and user interaction. The first wave of innovators is expected to be selected in November 2026. GOV.UK

Why does lifecycle oversight matter for CISOs and GRC teams?

The Commission's consultation revealed deep scepticism about current oversight: 65 per cent of Call for Evidence respondents disagreed or strongly disagreed that existing post-market surveillance for medical devices, including AI devices, was sufficient. The Commission specifically identified performance drift and context-dependent performance as reasons why pre-market testing alone may fail to predict real-world behaviour. GOV.UK

For enterprise security and governance leaders, this is a clear signal that static risk assessments are giving way to continuous monitoring. An AI model that passed validation in a controlled setting may still degrade when exposed to new data pipelines, edge-case patient populations or altered clinical workflows. This mirrors the shift from annual audits to real-time oversight, which is why many teams are investing in an AI governance dashboard that tracks model behaviour, version changes and drift indicators across the estate.

The change also expands the scope of accountability. Health Innovation Minister James Frith stated that "innovation must never come at the expense of patient safety," while HRA CEO Matt Westmore emphasised that maintaining public confidence will depend on preserving patient and public trust as AI moves into wider use. GOV.UK Health Research Authority

What are the risks and open questions?

Although the political commitment is firm, many of the concrete regulatory mechanisms are still to be designed. The government's response is an acceptance-and-implementation commitment, not the final regulatory rule itself. Legislation, licensing powers, detailed post-market requirements and change-control rules will follow in the Spring 2027 roadmap. GOV.UK

Capacity is a major unknown. Jennifer Dixon of the Health Foundation warned that the real test will be whether the NHS has the capacity, skills and systems to monitor AI safely at scale. Financial Times Enterprise teams should assume that similar resource constraints will apply inside their own organisations, because continuous oversight requires telemetry engineering, statistical monitoring and clinical governance expertise that many security and compliance functions do not yet possess.

The Commission's proposed "L-plates" model - described by Gerard Hanratty of Browne Jacobson as a staged-authorisation approach with close supervision and tight guardrails - also raises operational questions. Computer Weekly Organisations will need to define what provisional deployment looks like in practice, how to segment networks for partially authorised AI tools, and how to manage incident response when a device is operating under experimental conditions.

Finally, the Health Research Authority has been tasked with clarifying when AI development, deployment, evaluation and monitoring constitute research requiring research governance, including silent local evaluation and in-silico methods. This boundary between clinical care and research will affect data-consent regimes and ethics review obligations. Health Research Authority

What practical controls should enterprises adopt now?

Healthcare organisations should prepare for continuous evidence, monitoring and change-control obligations rather than treating AI assurance as a pre-deployment compliance exercise. There are five immediate priorities.

First, inventory every AI-enabled medical device currently in use and map its data dependencies, update mechanisms, training pipelines and current version-control practices. If an auditor asks for the exact version of a model that contributed to a specific patient record, the organisation must be able to answer.

Second, begin drafting internal Predetermined Change Control Plans now, even before MHRA publishes its December 2026 guidance. A PCCP should define which algorithmic updates are pre-approved within specified bounds, what testing is required for out-of-bound changes, and who has authority to approve them.

Third, build or procure telemetry that can detect performance drift and context-dependent degradation. The AI Airlock Phase 3 expects manufacturers to access deployment-specific data and define criteria for when safety or performance signals trigger action; buyers should demand the same transparency from vendors. GOV.UK

Fourth, integrate version identifiers and unique device identifiers into asset management systems. As MHRA explores linking version information to electronic patient records, enterprise configuration management databases must be able to support that traceability at scale.

Fifth, establish escalation thresholds for silent degradation. The new framework expects action when performance drops even if no reportable incident has occurred. Security and governance teams should define those thresholds with clinical stakeholders, document them in runbooks, and rehearse the response. Leaders should monitor the MHRA's December 2026 draft guidance and Spring 2027 roadmap through a weekly AI governance briefing to keep compliance programmes current as rules solidify.

What is the bigger picture for AI governance?

The UK's decision reflects a permanent philosophical shift: approval is becoming the start of an assurance lifecycle, not the end of regulatory scrutiny. The Commission gathered evidence from more than 12,000 patients, clinicians, healthcare leaders and developers, and its central conclusion was that regulation must become "more proportionate, lifecycle-based and system-wide." GOV.UK

For CISOs, CIOs and GRC leaders, the lesson extends beyond regulated medical devices. Any organisation deploying adaptive AI - whether in healthcare, finance, insurance or critical infrastructure - should assume that regulators will eventually demand the same continuous evidence of safety, version control and post-deployment monitoring. The firms that build those capabilities now will avoid a frantic compliance sprint when the Spring 2027 roadmap hardens into enforceable rules.

Frequently asked questions

What is the UK healthcare AI lifecycle oversight policy?

On 6 October 2026, the UK government accepted all 44 recommendations from the National Commission into the Regulation of AI in Healthcare. The policy shifts MHRA oversight from one-time pre-market approvals to continuous monitoring of AI-enabled medical devices throughout their working life, with a full implementation roadmap due by Spring 2027.

When will the new UK healthcare AI regulations take effect?

The government accepted the recommendations on 6 October 2026, but concrete rules are still being designed. MHRA plans to publish draft guidance on change control by December 2026, and the first full implementation roadmap covering all 44 recommendations is promised for Spring 2027. Final legislation and licensing powers will follow.

What are Predetermined Change Control Plans for healthcare AI?

Predetermined Change Control Plans, or PCCPs, are a proposed mechanism where manufacturers define in advance what changes an AI medical device can make automatically within specified bounds. MHRA accepted the recommendation to introduce PCCPs and committed to publishing draft guidance for public comment by December 2026.

How will MHRA monitor AI medical devices after deployment?

MHRA is moving toward tailored post-market surveillance, including ongoing performance reporting and escalation when performance degrades even without a reportable incident. Phase 3 of the AI Airlock sandbox is testing how to capture deployment-context factors such as local infrastructure, data quality and user interaction.

Sources

Part of our daily Briefing Analysis, built on the AI Governance & Security Briefing.

UK healthcare AIMHRAAI lifecycle oversightmedical device regulationAI governancepost-market surveillance
Share LinkedIn X

Get audit-ready before August 2026.

See how WalledAI operationalises Articles 9–15 and Article 50 obligations as a runtime governance layer.