The AI Airlock Phase 2 Programme Report has been published by the MHRA, providing information on the insights and subsequent recommendations from Phase 2 of the initiative.
Phase 2 builds upon the initial pilot programme launched in 2024, a regulatory sandbox used to address the challenges that come with regulating medical devices that incorporate artificial intelligence (AI).
Phase 2 worked with several candidates to focus on three specific areas in applying the current frameworks onto AI-enabled medical devices (AIaMDs) that pose difficulty, namely predetermined change control plans (PCCPs) and post-market surveillance, performance evaluation, and scope of intended purpose and validation, as these were originally designed with static devices in mind.
The report, published in June 2026, details five key insights, including the need to ensure devices, particularly those with generative AI and large language models, stay within their intended purpose. The Airlock highlighted how these devices, in the absence of guardrails, start to exhibit functionalities beyond those they were originally designed for. Other findings include the need for adequate human oversight, which was reflected in the National Commission’s call for evidence, and grounding the acceptance criteria and thresholds for PCCPs in clinical relevance.
Following the completion of Phase 2, recommendations were shared with the National Commission, whose recommendations to the Medicines and Healthcare products Regulatory Agency (MHRA) on evolving the current regulatory framework will likely impact the design of Phase 3 of the Airlock programme. The Phase 2 report findings have called for guidance development for PCCPs and performance metrics for AI-powered in vitro diagnostics (IVDs). There was also a recommendation to update the intended purpose and software qualification guidance to reflect the evolving device landscape. Considerations in the report were not limited solely to the MHRA, but also extended to manufacturers and approved bodies.
The Department for Health and Social Care (DHSC) has already given £3.6 million of funding to the AI Airlock to develop the programme and the regulatory sandbox concept further, aiming to tackle increasingly more complex regulatory challenges. The report detailed strong stakeholder feedback to the programme; however, it was noted that translating the findings into tangible guidance and policy changes will be crucial to its long-term success.
AI Airlock was not the only sandbox announcement from the MHRA last week, with Science Minister Lord Vallance announcing an AI Sandbox for medicines development during London Tech Week, supporting the delivery of safer medicines to patients. This new sandbox will allow companies and researchers to test AI tools in understanding the safety of a drug in a controlled environment. Work is expected to begin in the summer of 2026 between the regulator, industry and academia.
Furthermore, a joint initiative between the MHRA, NHS England and London Health Innovation Networks announced a London-specific regulatory sandbox, known as London Region I. Up to 10 AI medical device providers will be selected in the initial phase to work alongside the NHS in London to deploy technologies in live clinical settings, under the watch of the MHRA, with the aim of providing a controlled environment to generate real-world evidence on the safety and effectiveness of the technologies.
The three initiatives indicate the MHRA’s strong interest in allowing regulation to enable innovation, to work collaboratively with researchers and to evolve with changing technologies, a sentiment echoed by its Chief Executive, Lawrence Tallon, who, in response to the new medicines sandbox, stated, ‘by working directly with developers and researchers, we can help create the right environment for innovation to thrive and cement the UK’s position as a world leader in life sciences’.
