The Medicines and Healthcare products Regulatory Agency opened applications on Tuesday for the third phase of AI Airlock, its sandbox for testing how AI-enabled medical devices perform after they reach NHS wards.
Phase three focuses on post-market surveillance, implementing a recommendation from the National Commission into the Regulation of AI in Healthcare that devices should be monitored throughout their working life rather than judged on a single pre-market assessment.
Why lifecycle regulation matters
Machine-learning models can drift as patient populations change or hospitals update electronic health records. A tool cleared in 2024 might silently degrade by 2026 if training data no longer match real-world inputs. Ministers accepted all 44 commission recommendations on 6 October, committing to a more agile framework that still prioritises patient safety.
AI Airlock gives developers a structured environment to experiment with monitoring dashboards, human override workflows and incident logging before regulators embed requirements in statutory guidance. Themes for this phase were informed directly by commission conclusions.
NHS uptake context
Trusts already deploy AI for radiology triage, sepsis alerts and discharge planning, but boards want evidence that vendors will update models responsibly. NHS England’s commercial teams increasingly write lifecycle obligations into procurement contracts, mirroring what Airlock tests in miniature.
Clinicians worry about alert fatigue if post-market surveillance generates noise without actionable thresholds. Sandbox participants must show how hospitals would review flags without pulling nurses from bedside care.
Patient trust and transparency
Patient groups welcomed the government’s response but demanded plain-language notices when AI contributes to a diagnosis. Airlock phase three experiments with patient-facing summaries and audit trails that could become standard labelling elements.
International partners watch UK sandboxes because MHRA decisions influence mutual recognition talks post-Brexit. A credible post-market regime could speed access for British startups if EU manufacturers face slower updates under MDR paperwork.
Commercial incentives
Developers pay staff time to join sandboxes, betting that early engagement reduces costly rework later. Venture investors say lifecycle monitoring costs must be baked into business models, not treated as compliance afterthoughts once NHS contracts are signed.
Larger medtech firms may use Airlock to compare internal monitoring tools against regulator expectations, especially where devices combine software and hardware updates on different cadences.
Next regulatory steps
MHRA’s 2030 strategy promises an AI-enabled clinical trial protocol checker and clearer digital health expectations. Phase three outputs will feed those guidance documents. Hospitals should expect formal post-market plans to become mandatory for high-risk AI devices within the parliament.
For patients, the practical benefit is safer software that improves over time with evidence, rather than frozen algorithms cleared once and never reviewed. Tuesday’s opening is bureaucratic, but it marks the shift from pilot rhetoric to operational oversight.
