MHRA opens third AI Airlock phase with post-market surveillance as first focus

The Medicines and Healthcare products Regulatory Agency (MHRA) has opened the third phase of AI Airlock, its regulatory sandbox for AI-enabled medical devices, and made post-market surveillance across the device lifecycle its first priority ¹². Applications opened on 6 October 2026, and manufacturers that want a place in the first review wave must apply before 31 October ².

What the MHRA is asking for

The call sets out three challenge areas ². The first is how manufacturers can design monitoring that captures the local factors that change how a device behaves once deployed, including infrastructure, workflow integration, data quality and how staff use it. The second is how to spot and respond to safety or performance signals where clinical oversight is thin, such as remote, direct-to-consumer or more autonomous systems. The third is how information on performance, updates and rollbacks reaches regulators, providers, professionals and patients in time ².

The application window stays open all year, with candidates reviewed in waves and onboarded on a rolling basis; a further shortlisting period is planned for spring 2027 ². Phase 3 is funded until April 2029, and the MHRA expects each project's testing to run for 6 to 12 months ². Projects can be tested in a simulation workshop, a virtual testing environment using real or synthetic data, or in the real clinical setting where the product would be used ².

In a blog published on 19 October, the agency explained the choice ¹. Its pilot looked at continuous, near-real-time monitoring through Newton's Tree's FAMOS platform at 2 deployment sites, covering data quality, performance drift and automation bias ¹. Phase 2 concluded that validation at a single point before market entry may not show how an AI device will perform in practice, and linked monitoring to predetermined change control plans, which set out in advance how software changes will be assessed ¹.

What it means for the market

For AI device makers selling into the NHS, the direction is clear. The MHRA describes monitoring as something to be designed early, alongside pre-market evidence and lifecycle plans, rather than added after launch ¹. Suppliers that can show how they will set thresholds for action, detect drift and decide on rollbacks are likely to find themselves closer to where the regulator is heading.

The blog also points to buyers. It names healthcare organisations and professionals, as well as manufacturers and regulators, as sharing responsibility for spotting and managing risk once a product is in use, and says the respective roles may need further clarity ¹. For NHS trusts deploying AI tools, that suggests closer scrutiny of what local data they provide to suppliers and who reviews the signals.

The MHRA ties the work to the National Commission into the Regulation of AI in Healthcare, which called for regulation to cover the whole lifecycle ¹. Multi-year funding from DHSC lets the programme move beyond annual budgets and run longer tests ¹. Outputs such as regulatory tools, frameworks and best practice are meant to be published for the wider sector ².

The agency stresses limits on what it is trying to settle. It says it is not setting out to fix a single solution or policy position, but to gather evidence on which approaches work and where existing rules suffice or further guidance is needed ¹. It also warns against monitoring that is impractical or produces large volumes of data that do not support decisions ¹.

Caveats

The sources describe a call for applications and the agency's reasoning, not results. Neither names the companies that will take part or how many places are available, and the MHRA will take a portfolio approach to selection, favouring projects that maximise learning ². The first selection wave closes to applicants on 31 October 2026 ².