AI Airlock Phase 2 Cohort
Phase 2 of the Airlock will include seven additional technologies spanning AI-powered clinical note taking, advanced cancer diagnostics, eye disease detection tools, and obesity treatment support systems.
Documents
Details
Overview
Phase 2 of the Airlock will include seven additional technologies spanning AI-powered clinical note taking, advanced cancer diagnostics, eye disease detection tools, and obesity treatment support systems. This expansion addresses three fundamental regulatory challenges around managing evolving AI applications, regulating AI-powered diagnostics effectively, and implementing robust post-market surveillance for AI medical devices.
Candidates will test in the Airlock until March 2026 when the second phase will be finalised.
The phase 2 sandbox structure
Phase 2 of the AI Airlock is structured around 3 key regulatory challenge areas. Within each challenge there are multi-environment candidates and simulation candidates.
The multi environment candidates will work with the AI Airlock team to test the regulatory challenge in more than one of the AI Airlock’s testing environments:
Simulation environment: a focused roundtable workshop bringing together multiple stakeholder perspectives to address simulated testing scenarios
Virtual or Research environment: testing of the AI product in a controlled data environment to address specific research questions and generate evidence
Real-world environment: deployment of a product in the intended use environment, separate from the clinical pathways and decision-making processes, to generate evidence of the model’s deployment without impacting the healthcare outcomes
The Simulation environment candidates will work with the Airlock at the Simulation level only, helping to bring multiple perspectives to the challenge by varying the product type, clinical area and development stage of the case studies under investigation.
Regulatory challenge 1: Scope of intended use extension and validation
Multi-environment candidate:
TORTUS
TORTUS is an evolving clinical AI assistant designed to support healthcare professionals by reducing administrative burden and improving quality of care. Today, TORTUS provides automated clinical documentation, integrated directly within electronic health records.
A central regulatory challenge for technologies like TORTUS is defining the boundary where features move from documentation support into diagnostic or decision-support functionality. Under the UK Medical Device Regulations, such a change in intended use could result in reclassification from Class I to Class IIa. While clear thresholds exist for traditional medical devices, there is no established benchmark for AI-based clinical documentation systems, creating uncertainty for both innovators and regulators.
TORTUS has been selected into Phase 2 of the MHRA AI Airlock programme to address this challenge, with a focus on intended use and validation. The project will explore practical approaches to verification, validation, and post-market surveillance for LLM-enabled functionality operating at higher regulatory thresholds. By working closely with regulators and clinical stakeholders, TORTUS aims to help shape a clearer framework that safeguards patient safety and compliance, while enabling innovation in clinical AI to progress responsibly and effectively.
Simulation candidate:
Nu & Aegis AI Conversational & Monitoring System - Numan
Blood test results are a key source of health information but are often difficult for patients to interpret accurately. Many seek informal explanations online, where information quality is variable and can present safety risks.
Numan’s AI system aims to address this gap by providing personalised, evidence-based explanations of blood test results in accessible language. The regulatory challenge focuses on how such AI systems can offer explanatory support to patients in a way that is demonstrably safe, transparent, and validated.
Initially, the AI will be limited to providing explanatory information. With further evidence and oversight, its functionality could expand to include result summaries and personalised insights, within clearly defined safety parameters.
Through the AI Airlock, Numan will test how this direct-to-patient model aligns with existing regulatory frameworks, contributing to a clearer understanding of how similar innovations can be safely deployed in the future.
Regulatory challenge 2: AI-powered in-vitro diagnostic devices
Multi-environment candidate:
PANProfiler Colorectal (MSI/MMR) - Panakeia Technologies
Around 44,100 people in the UK are diagnosed with colorectal cancer each year, yet only 58% begin treatment within the NHS-recommended 62-day window. Treatment delays often arise because patients require additional laboratory tests for mismatch repair deficiency (MMR/MSI) which help doctors decide on the best treatment. These tests typically take 2-4 weeks, during which patients cannot start treatment, which causes stress and delays care. Patients with MMR deficiency can benefit from immunotherapy, which can increase survival rates by up to 80%.
PANProfiler Colorectal (MMR/MSI) uses artificial intelligence to analyse the same tissue slides pathologists use for diagnosis. The AI delivers results within minutes, without extra tests, allowing doctors to make faster treatment decisions.
The technology has been validated on over 4,700 UK colorectal cancer samples, showing similar accuracy to current laboratory tests, and it has now been approved for NHS use.
Regulatory challenges remain: laboratory methods vary across hospitals, tumours can behave differently, and definitions of “positive” or “negative” results are not fully standardised. Panakeia is working closely with the MHRA, NHS, and experts to set clear standards, ensuring AI tools are safe, reliable, and ready for widespread NHS adoption.
Simulation candidate:
Octopath
The UK faces a pathology crisis that directly impacts cancer patient outcomes. Advances in personalised Medicine require increasingly complex, time-consuming diagnostic analysis to select targeted therapies. This demand is colliding with a shrinking specialist workforce, a 25% shortfall in staff, and high retirement rates. This bottleneck creates unsustainable workloads and diagnostic delays. In March 2024, over 58,000 people in England waited over four weeks for a cancer diagnosis. There is a clear, unmet need for a solution that enhances diagnostic efficiency and ensures patients benefit from precision oncology.
Octopath addresses this by augmenting pathologists with a state-of-the-art AI platform, delivering rapid, quantitative analysis from routine digital pathology slides. Our models provide clinically relevant outputs like mitotic index and immune cell infiltration to support faster, more consistent diagnoses. However, as an adaptive AI in-vitro diagnostic device, Octopath presents novel regulatory challenges, particularly in clinical evaluation and managing post-market updates.
Participation in the AI Airlock is critical to collaboratively address these challenges. By working with the MHRA explore approaches to generating clinical evidence and a Predetermined Change Control Plan, we aim to provide a practical case study to inform the UK’s regulatory framework, helping to create a clearer, evidence-based pathway for future innovators
Regulatory challenge 1: Post-market Surveillance and Predetermined Change Control Plans
Multi-environment candidate:
NHS England Federated Data Platform – Safe Summarisation
The Safe Summarisation tool will use Large Language Models (LLMs) to produce coherent summaries of a patient’s hospital stay. Drawing from multiple sources of clinical notes it will generate a single accurate summary. These can be tailored for clinicians briefing colleagues or for patients and their families seeking a simplified overview of the care administered during their stay. The tool may also support the creation of discharge summaries and assist with clinical coding.
A key aim of the project is to explore the regulatory factors of applying AI in this way. This includes demonstrating safety for multiple use cases and enabling live monitoring and updates by hosting the tool on the NHS Federated Data Platform (FDP).
The NHS FDP provides a secure framework for managing AI medical devices. It allows data to be analysed locally within NHS Trusts, supporting robust Post-Market Surveillance (PMS) to proactively monitor for performance issues and safety in near real-time. This creates a proactive, evidence-led approach to AI governance, ensuring the tool remains safe and effective.
Simulation candidate:
Eye2Gene – Eye2Gene Ltd (and Hardian Health)
Eye2Gene™ is an expert AI as a medical device product that informs the diagnosis of genetic eye disease from a patient’s retinal scan to help more patients get a genetic diagnosis sooner. Eye2Gene™ addresses a critical unmet need: timely, equitable access to expert-level diagnosis, which is currently limited to a small number of specialist centres.
Effective Post-Market Surveillance (PMS) is crucial for Eye2Gene™, requiring continuous monitoring of real-world performance, bias detection, and data drift, which can be linked to variation in retinal scan technology, retinal image quality, population level differences and novel genetic discoveries. This needs infrastructure for automated logging, outcome capture, and clinician feedback loops.
The adoption of Predetermined Change Control Plans (PCCPs) by Eye2Gene™ will define allowable updates, validation, and re-submission that will be key to balancing extensions and recalibration of the AI with regulatory conformity for patient benefit.
DeepXAI – DeepX Health UK Ltd
DeepX Health has developed DeepX AI to support early skin cancer detection. DeepX AI analyses dermoscopic images of skin lesions to identify lesion types and provide triage recommendations to clinicians. It is designed to integrate into existing clinical workflows and enhance them by increasing sensitivity to potential malignancies, reducing unnecessary referrals, and improving triage efficiency. While DeepX Health’s core imaging platform is already cleared in the US and EU and is in clinical use, the AI component is not yet UKCA-marked.
A key regulatory challenge for AI-based medical tools is enabling performance improvements and broader clinical use without triggering repeated full conformity assessments.
Through Airlock, DeepX Health seeks to explore how robust post-market surveillance (PMS) and pre-determined change control plans (PCCPs) can support responsible, real-world evolution of medical AI products. We aim to co-develop practical, evidence-based approaches to AI oversight that ensure safety and equity across patient populations, while allowing for technological and clinical advances. The goal is to reduce regulatory uncertainty, contribute to future regulatory guidance on AI as medical devices, and enable scalable yet responsible deployment.
AI Airlock Partners
The AI Airlock Partners comprise of the following organisations:
- Medicines and Healthcare products Regulatory Agency
- Department of Health and Social Care
- NHS AI Team
- Team AB
The ICO will also be supporting the MHRA AI Airlock via a referral service offering data protection by design advice to applicants. If you would like support, please indicate this during the application process.
Fees & Funding
While there is no fee for the applicants, throughout the Airlock programme. Upon joining the Airlock there will be a resource commitment to be made between the candidates and the MHRA. Candidates are expected to fund their own studies and delivery of any Airlock testing, including accessing relevant data sets.
Phase 2 testing timelines
The second phase of the AI Airlock will run until April 2026. While each sandbox testing plan will be bespoke to the product, candidates should expect to complete their individual Airlock testing within 6 months. This timeframe is aligned with emerging global best practices.
Further information
If you would like to ask any questions, please email aiairlock@mhra.gov.uk.