Health regulators opened the next phase of the MHRA’s AI Airlock programme on the same day the government pledged to implement every recommendation from a year-long NHS-led review of how artificial intelligence should be supervised in clinics and hospitals.

Commission outcomes

The National Commission on the Regulation of AI in Healthcare delivered 44 recommendations after engaging more than 12,000 people across the UK. Ministers confirmed on 6 October that they would accept the full set, including lifecycle oversight, post-market monitoring and clearer classification of software that adapts after deployment.

A programme board will track delivery, with a public roadmap due by spring 2027. Short-term commitments include draft MHRA guidance on managing changes to AI-enabled devices as they learn from new data, targeted for December 2026.

AI Airlock phase three

The sandbox lets developers test promising tools under regulatory supervision before wider NHS rollout. Applications for the third phase close in November 2026, with a webinar for prospective participants on 22 October. Themes reflect commission priorities such as staged authorisation pathways that allow supervised use while real-world evidence accumulates.

Manufacturers have long complained that point-in-time approvals struggle to keep pace with models that update frequently. The government’s response explicitly contemplates lifecycle regulation paired with stronger post-market surveillance.

What clinicians will notice

Staged pathways could let trusts pilot diagnostics or triage assistants under tighter monitoring than a full marketing authorisation would normally allow. That may speed access for cancer or emergency pathways, but it also shifts responsibility onto local governance committees to document outcomes.

Devolved health administrations are included in the implementation plan, meaning Wales, Scotland and Northern Ireland will need aligned reporting even where procurement differs.

Timeline and risks

Consultations on qualifying and classifying AI devices are slated to begin next year. Until then, existing medical-device rules remain in force, and trusts should treat commission language as direction of travel rather than immediate licence to deploy unapproved models.

The policy bet is that Britain can keep patients safe without copying every EU AI Act process—a claim that will be tested the first time a sandbox product fails in live care.

Industry questions

Small med-tech firms worry that sandbox slots favour well-lawyered multinationals. The MHRA has promised webinars to demystify applications, yet legal costs for a single submission can still exceed a start-up’s annual budget without grant support.

Editors will update this file if ministers publish new data later in the day.