MHRA has refreshed its UK guidance for software and artificial intelligence (AI) as a medical device, with direct relevance for pharmacovigilance and device vigilance teams supporting AI-enabled medical devices.
The MHRA “Software and artificial intelligence (AI) as a medical device” guidance page is marked as updated on 3 February 2025. In that update, MHRA highlights internationally aligned work undertaken with the U.S. Food and Drug Administration (FDA) and Health Canada, including jointly identified five guiding principles for the use of predetermined change control plans (PCCPs) for machine-learning medical devices, and jointly established guiding principles on the transparency of machine-learning medical devices.
While the fact sheet excerpts do not include the full text of the joint principles, the MHRA page’s signposting is notable for teams that must evidence lifecycle governance and post-market oversight for adaptive software.
In parallel, MHRA’s Software and AI as a Medical Device Change Programme roadmap—marked as updated 14 June 2023—continues to provide a structured view of MHRA’s lifecycle approach. The roadmap lists work packages spanning qualification and classification, premarket requirements, post-market, cyber secure medical devices, and topics on AI interpretability and AI adaptivity.
For organizations placing SaMD/AIaMD on the UK market, the combination of MHRA’s signposting to PCCP and transparency principles and the lifecycle roadmap’s emphasis on post-market, cybersecurity, interpretability and adaptivity provides a practical cue: governance and evidence packages should be organized so that device performance changes—especially those associated with algorithm updates—are managed under controlled change processes and linked to post-market monitoring.
This matters because adaptive ML medical devices may be updated post-deployment (or may change behavior depending on their design), and safety-system owners may need to demonstrate that updates are anticipated, documented, and monitored with clearly defined escalation criteria.
As a conditional next step, if your organization fields or updates SaMD/AIaMD that uses adaptive ML, safety/device vigilance and PV CSV leads can (1) confirm whether any marketed/fielded SaMD/AIaMD uses adaptive ML or planned updates that would fit a PCCP-like approach; (2) review current change-control SOPs and technical documentation to ensure change governance and transparency artifacts are captured and retrievable; and (3) verify post-market/vigilance processes explicitly account for algorithm updates, including monitoring after updates and escalation criteria.
No explicit compliance deadline is stated in the cited sources; this governance check can be incorporated into the next AI governance or device PMS review cycle. The provided excerpts do not include the detailed content of the joint PCCP or transparency principles; teams may need to review the underlying MHRA page and linked materials directly when updating internal procedures.
