AI Regulation Tracker / Regulator sandbox
UK MHRA expands its AI Airlock sandbox, and its findings will shape AI medical-device compliance expectations
Britain's medicines regulator has funded a three-year run of its AI Airlock, the regulatory sandbox for AI as a medical device. The tests it runs now will shape the compliance rules that AI medtech firms and the hospitals buying their tools face later.
The UK's Medicines and Healthcare products Regulatory Agency has committed 3.6 million pounds over three years to expand the AI Airlock, its regulatory sandbox for artificial intelligence used as a medical device. The MHRA announced the funding on April 8, 2026. The money comes from the Department of Health and Social Care and amounts to about 1.2 million pounds a year from 2026 through 2029.
The AI Airlock is not a law. It is a controlled testing environment, first launched in spring 2024, where the regulator and a small set of companies work through how existing UK medical-device rules apply to AI tools that learn, update, and sometimes behave unpredictably. For any professional buying or building this technology, the significance is not the pound figure. It is that the standards being worked out inside the sandbox are the standards the market will be measured against later.
What the sandbox is testing
The MHRA describes the Airlock as the UK's first regulatory sandbox for AI as a medical device, run in partnership with the DHSC, the NHS AI team, and the agency's approved-body partners. According to the MHRA, the second phase explored a range of technologies, including large language models, voice tools, and diagnostics for cancer and rare diseases.
The regulator says the work centers on the questions that make AI devices hard to regulate with older frameworks. Those include risk management specific to AI, such as grounding a model's outputs in verified clinical information to cut errors, the explainability of AI recommendations so clinicians can trust them, and ongoing post-market monitoring for both performance drift and user over-reliance. The programme is also examining pre-determined change control plans, the mechanism that lets a device be updated after approval without a full new review, and how a product's approved scope can expand.
James Pound, an MHRA executive director, said the investment "will allow us to scale up and ultimately strengthen our ability to ensure that AI-powered medical devices can reach patients safely and efficiently," with the aim of confident regulatory oversight.
Why a preview matters more than a rule
The reason a sandbox earns attention from compliance teams is that it front-runs formal rulemaking. What the MHRA learns about how to classify the risk of a diagnostic model, how to police a change-management plan, and how to hold a device to a fairness standard tends to reappear later as guidance, and eventually as the expectation an approved body applies at market entry.
The MHRA says the insights from the Airlock are feeding the National AI Commission's work on the future regulation of AI in healthcare. That is the channel through which sandbox findings can harden into settled policy. A hospital signing a multi-year contract for an AI triage tool, or a developer designing a device today, has a practical interest in reading those findings early rather than discovering the expectation at inspection.
What it does not do
The announcement does not create a new binding rule, a new penalty, or a new approval requirement. Taking part in the Airlock does not grant a device market authorization, and staying outside it does not put a company in breach of anything. The programme is guidance-forming. Its output is knowledge and, in time, guidance, not an enforceable obligation on the day it is published. Firms should be careful not to read the summer 2026 report as a rulebook. It is a signal of direction.
For a US reader, the Airlock is worth watching as a parallel to the FDA's evolving approach to AI and machine-learning-based devices, including its own work on predetermined change control plans. A medtech company marketing on both sides of the Atlantic will need to map the UK route against the US one, because the two regulators are circling many of the same questions, risk classification, post-market monitoring, and change management, without yet converging on identical answers.
The near-term hook is the Phase 2 report. The MHRA says reporting is expected in summer 2026. That document, more than the funding line, is what buyers and builders of AI medical devices should read when it lands.
Frequently Asked Questions
What exactly did the MHRA announce?
On April 8, 2026, the MHRA announced 3.6 million pounds in Department of Health and Social Care funding, about 1.2 million per year from 2026 to 2029, to expand the AI Airlock, its regulatory sandbox for AI as a medical device.
Who does this affect?
Companies developing AI-based medical devices for the UK market, and the NHS trusts, hospitals, and clinicians that buy and rely on those tools. The sandbox is where UK medical-device rules are tested against high-risk AI.
Is the AI Airlock a binding regulation?
No. It is a regulatory sandbox, meaning a supervised testing environment. It is guidance-forming, not a binding rule. It creates no new penalty or approval requirement, but its findings preview the compliance expectations the MHRA is likely to apply later.
When are the findings due?
The MHRA says the Phase 2 report is expected in summer 2026. It is set to cover risk classification, change-management plans, and bias and fairness questions for AI medical devices.
Sponsored Training
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.