Key Takeaways
- The National Commission into the Regulation of AI in Healthcare, established by the UK’s MHRA, recommends a “staged authorisation” system by September 2026 for new AI medical devices, modelled on provisional “L-plate” licences, requiring manufacturers to prove real-world safety before full approval.
- Mandatory continuous post-market monitoring addresses model drift, the risk that an AI system’s behaviour changes as it processes more real-world data, and is central to the Commission’s case for giving patients earlier access without relaxing long-term safety standards.
- A May 2026 King’s College London study found that three-quarters (76%) of the UK public want AI medical tools officially regulated even if that delays adoption, while a Royal College of Physicians survey found 73% of doctors cite algorithmic error as their primary concern with clinical AI. Three-quarters of the UK public say AI tools in patient care should be officially approved and regulated, even if that slows their rollout with only 17% saying doctors should be free to choose AI tools without official approval. That finding, from a May 2026 King’s College London study, frames the challenge facing the National Commission into the Regulation of AI in Healthcare, which published its blueprint on September 10, 2026, proposing “L-plate” provisional approvals for AI medical devices and mandatory real-world monitoring once they are deployed.
The L-Plate Approach
The Commission was established by the Medicines and Healthcare products Regulatory Agency (MHRA) and gathered evidence from more than 12,000 patients, clinicians and industry leaders before issuing its report. Its core mission: protect patients, support the government’s ambition to make the NHS a leading AI-enabled health system, and build a regulatory environment that attracts innovators without compromising safety. According to the MHRA, Lawrence Tallon, its chief executive, said the UK aims to be the best place for developers to build and test AI, for clinicians to use it and for patients to benefit from it.
The staged authorisation model is the Commission’s sharpest proposal. Under the L-plate framework, a new AI model would be deployed initially under close supervision with tight guardrails, giving manufacturers the opportunity to demonstrate real-world safety before receiving full authorisation. The approach acknowledges a fundamental difference between AI and traditional medical devices: AI continues to adapt after licensing as it processes more data. That adaptivity is the source of both its clinical promise and its principal regulatory risk. An AI system can drift, gradually ceasing to behave as intended, in ways that a static device cannot. Staged authorisation is designed to give regulators the visibility to catch that drift before it reaches patients at scale.
Post-Market Monitoring
Continuous real-world monitoring throughout a device’s operational life is a companion recommendation to staged authorisation, not an afterthought. The Commission’s position is that initial approval is not a destination; it is the beginning of a regulated lifecycle. Stephen Harden, president of the Royal College of Radiologists specifically welcomed the monitoring provisions, noting the increasing complexity of AI in clinical settings.
The Care Quality Commission (CQC), which regulates health and social care services in England, reinforces this ongoing scrutiny at the service level. The CQC does not approve individual technologies, but its guidelines make clear that AI should enhance rather than replace human decision-making, with continuous human oversight of AI outputs. Transparency requirements apply too: patients must receive adequate information about AI’s role in their care. The CQC also identifies cybersecurity, fairness across population groups and staff training as compliance obligations for services deploying AI tools.
What Clinicians and Patients Say
The same King’s College London study found that AI chatbots misdiagnose in up to 80% of early medical cases, though that figure should be treated with caution given the variability of testing methodologies across sources.
Royal College of Physicians survey found that 73% of doctors cite algorithmic error as their primary concern with clinical AI, with many also reporting low confidence in using the tools and a need for more training Those numbers suggest the regulatory framework the Commission is building needs to address professional readiness as much as device safety, two problems with different solutions.
The Wider Regulatory Architecture
The Commission’s recommendations land inside a broader MHRA programme already in motion. A dedicated framework for AI as a medical device is expected later in 2026, aligned with the UK government’s Life Sciences Sector Plan, and will introduce structured requirements for AI lifecycle governance, transparency and cybersecurity. New post-market surveillance requirements came into force in June 2025 under the Software and AI as a Medical Device Change Programme. The AI Airlock pilot, a regulatory sandbox for AI medical devices, completed in March 2025 and is feeding directly into the framework’s development.
Sitting across all of this is the AI and Digital Regulations Service, a joint effort involving the MHRA, the National Institute for Health and Care Excellence (NICE), the Health Research Authority and the CQC. Its purpose is to give developers and adopters a single point of regulatory orientation rather than four separate compliance tracks. Whether that coordination holds as each body develops its own AI-specific guidance is a practical question the Commission’s report does not fully resolve. The fragmentation problem is not uniquely British; governments building AI regulatory architectures across multiple agencies face the same coordination risk.
Originally published at https://autonainews.com/uk-proposes-l-plate-ai-approvals-for-medicine-with-mandatory-monitoring/
Top comments (0)