The government has accepted all 44 healthcare AI commission recommendations, and MHRA has opened AI Airlock Phase 3. The sandbox is not funding, product approval or evidence that proposed rules are already in force.
What changed on 6 October
The UK government has accepted all 44 recommendations of the National Commission into the Regulation of AI in Healthcare. Its response sets out an action plan and delivery priorities. This materially updates the commission report covered here in September: there is now a government commitment, rather than advice awaiting a response. Acceptance does not itself put every proposed rule into force or establish the safety of a particular product.
AI Airlock Phase 3 is open—but it is not a grant or approval route
MHRA's Phase 3 application window opened on 6 October and is intended to remain open year-round until the official page says it has closed. Applications are reviewed in scheduled waves and onboarded on a rolling basis; the current wave closes on 31 October, with a later shortlisting period planned for Spring 2027. Organisations of all sizes may apply, and a lead applicant may partner with other organisations. The present priority is post-market surveillance and lifecycle oversight for AI-enabled medical devices, although other regulatory challenges can be submitted as expressions of interest. Application and participation carry no fee, but selected teams are expected to fund their own studies, delivery and relevant data access. HubAI classification: regulatory sandbox support—not a grant, certification, conformity assessment or market authorisation.
What an applicant should be ready to show
MHRA says candidates may be asked for evidence of the idea's basic principle, relevant certification and regulatory or technical and clinical documentation, and data-access details where relevant. A proposal should identify patient or health-system benefit, the regulatory challenge, a testable hypothesis and the resources needed for a bespoke plan. Engagement is generally expected to last 6 to 12 months, while the current Airlock phase is funded until April 2029. The programme page promises collaboration and learning, not a successful regulatory outcome. A prospective applicant webinar is scheduled for 22 October 2026 at 10:00, and the first wave is expected to be selected in November.
The next milestones are drafts and consultation
The MHRA intends to issue draft guidance on Predetermined Change Control Plans for public comment by December 2026. A consultation on software and AI device classification is intended by Spring 2027. The government also plans an implementation roadmap by Spring 2027 and annual progress updates thereafter. These are stated delivery intentions, not completed guidance or a guaranteed commencement date for a new regime. HubAI recommendation: record each milestone as pending until its official document appears.
Why a launch-time assessment is still incomplete
The commission report was published on 10 September 2026. Its central procurement problem remains useful: a system can change after the initial assessment. HubAI analysis: the purchase should include an owned process for checking changes, responding to incidents and deciding whether to pause use. A demonstration cannot tell a buyer whether a later model update, new integration or different patient population will behave identically. Compare that operating arrangement alongside the licence price.
What healthcare providers are being told
The MHRA's accompanying provider explainer describes better product evidence, clearer deployment responsibilities, readiness support, role-specific training and help with monitoring, updates and reporting. It describes intended support as implementation develops, not services or outcomes already delivered to every provider. A procurement team can turn those areas into evidence questions now: who reviews the tool, who trains staff, who receives incident reports and who owns the system after launch?
Put the update boundary into the buying decision
HubAI recommendation: ask a shortlisted supplier to identify the intended use, excluded uses, product and underlying model versions, permitted integrations, update notice and rollback route. Require a named owner for re-review when any of those elements changes. Write down which evidence is supplier-reported, locally observed or unavailable. Ask how the contract handles a disputed output, a broken dependency or a material update. These are procurement questions, not a substitute for a qualified regulatory assessment.
A practical evidence register
Maintain one dated record per workflow: the version tested, authorised data, acceptance rule, corrections, unresolved risks, reviewer and next review trigger. Keep administrative drafting separate from work that could influence clinical decisions. Clinical deployment requires qualified clinical, regulatory and information-governance assessment; a HubAI checklist cannot provide that approval. Avoid copying a generic badge into the register as if it proved suitability for a particular use.
HubAI buyer verdict
Prepare the evidence and change-management process now; do not buy or certify a product solely because government has accepted the recommendations or a supplier joins the Airlock. Eligible developers with a defined lifecycle-monitoring challenge should assess the 31 October review wave, including the 6-to-12-month resource commitment and self-funded testing costs. Before expanding a pilot, name the person who can stop it and agree how a changed version returns to review. Revisit the plan when the December draft or Spring roadmap is actually published. No saving, clinical benefit or regulatory approval is guaranteed.
Update record and evidence limits
This existing briefing was first published on 14 September 2026 and updated on 6 October after the government response. Independent editorial coverage; not sponsored. Official commitments are distinguished from HubAI procurement analysis and completed implementation. No new product, comparison or HubAI Score is created. The accessible first 25 UK Trends results contained no directly related healthcare AI query; no search-volume or rising-interest claim is made. The same topic-specific conceptual cover is retained for this same article; it does not depict a clinical trial or approved product.
HUBAI VIEWBuild change notification, monitoring ownership and qualified review into procurement now; eligible medical-device teams can assess the Airlock route without confusing participation with approval.
Buyer decision signal: Government response · accepted recommendations, implementation pending
What to verify next
1Record the government response separately from current rules and future drafts
2For the current Airlock review wave, decide and submit before 31 October 2026
3Budget applicant-funded studies, delivery and data access; do not book the sandbox as grant income
4Prepare the regulatory challenge, testable hypothesis, evidence and 6-to-12-month resource owner
5Watch for MHRA draft change-control guidance by December 2026
6Track the classification consultation and implementation roadmap planned by Spring 2027
7Document intended use, exclusions, model dependencies and exact tested version
8Name the qualified reviewer, monitoring owner and person who can pause use
9Agree supplier update notice, re-review triggers, incident support and rollback
10Budget for staff training, monitoring and review alongside licence cost
Read the evidence
Capabilities, availability and prices can change. HubAI keeps analysis separate from the underlying official material.
01GOV.UK: government response and action plan — 6 October 2026Open source ↗02MHRA: AI Airlock Phase 3 application and eligibility — 6 October 2026Open source ↗03GOV.UK: government announcement and applicant webinar — 6 October 2026Open source ↗04MHRA: healthcare providers explainer — 6 October 2026Open source ↗05GOV.UK: original commission recommendations — 10 September 2026Open source ↗
Regulation
Safety