The world's first regulatory sandbox for AI medical devices just got a 3-year extension, and £3.6 million in funding to go with it.
It is a busy month for AI as a Medical Device (AIaMD) regulation. After the IMDRF's new consultation, the MHRA announces the extension of its AI Airlock programme.
The programme is the only place in the world where AI medical devices go through live regulatory scrutiny — with MHRA, the NHS AI Team, UK Approved Bodies, and DHSC all at the table — specifically to figure out how existing rules hold up against AI and where they break.
Here is what two phases of testing have surfaced for manufacturers.
LLMs are in the sandbox now
Phase 2 tested large language models, voice tools, and cancer and rare disease diagnostics. If you are building with foundation models in a clinical setting, MHRA is actively figuring out how to regulate you — not in theory, but on live products.
Post-market monitoring of AI drift is now a named priority
The pilot identified that AI systems can degrade in ways traditional devices do not: model performance shifts, user over-reliance, scope creep. MHRA expects you to catch it. Your post-market surveillance plan needs to account for this.
Pre-determined change control plans (PCCPs) are being stress-tested
How do you handle an AI device that evolves? Phase 2 is specifically exploring change control plans for AI-powered diagnostics.
Transparency is a design requirement
The pilot made clear that clinician confidence depends on AI transparency. If your device cannot explain its reasoning in a way clinicians trust, that is a regulatory problem.
The multi-year funding means MHRA can now run longer, more ambitious testing cycles than the earlier phases allowed.
Phase 3 applications have not opened yet. If you are building AI-enabled medical devices, this is the time to start watching.