Day 2 - Tuesday 16th JuneTuesday, 16 June 2026 - CEST Time Zone

• MHRA's AI Airlock launched in 2024, is the UK’s first regulatory sandbox for AI as a Medical Device (AIaMD). It provides a safe, collaborative environment for testing real-world AI medical products against regulatory challenges, enabling developers, regulators, and partners to co-create solutions that inform future guidance and policy.
• This presentation will share insights from the pilot programme including establishment of the sandbox, key milestones and regulatory challenges investigated and lessons learned. The programme is now in its second phase and key details about the candidates and products selected for Airlock will be provided, including the criteria for their selection. We will share important learnings from the candidates’ experiences—what worked well, common challenges faced, and successes achieved.
• Finally, the presentation will address future plans for Airlock, focusing on how the programme enables an innovative approach to regulatory policy, emphasise the importance of continuous engagement as a regulator and why designing regulations with real world input is effective.

• Methodology presentation: Establishing comparative performance benchmarks for multiple generic and domain-specific large language models (LLMs) for SLRs and by testing a range of datasets, including publicly available SLRs and anonymized contributor datasets for SLRs representing different medical device product categories and therapeutic fields.
• Testing results: Benefits in terms of both efficiency and accuracy - reduce reviewer SLR completion times over manual methods while improving LLM model performance and data/evidence quality by 7-25%. By focusing on user-friendliness, this research project also aims to reduce the time to train intermediate staff on AI. On a wider scale, the project anticipates a dramatic reduction in the time taken for industry standard adoption.
• Achieving Notified Body Acceptability: Demonstrate that the output of AI-driven data extraction—specifically the automated detection of longitudinal adverse events—is supported by a transparent, auditable data provenance trail that satisfies regulatory scrutiny under the EU MDR/IVDR.
• Defining the Role of Human Oversight: Establish best practices for the "human-in-the-loop" verification process, outlining when and how subject matter experts must validate LLM outputs to maintain regulatory compliance and manage risk.
• Validating AI Accuracy: Address the industry hurdle of accepting that a validated LLM model can achieve higher accuracy and lower error rates than manual human review, driving the 7-25% improvement in data quality

• Exploring the regulatory challenges and requirements for managing risks associated with AI-based Medical Device Software (AI-MDSW) under the EU AI Act.
• Discussing practical strategies for integrating AI risk management into existing regulatory frameworks to ensure compliance and patient safety.

Gap analysis of existing biological evaluations under the new ISO 10993-1 revision

Workshop Summary: This interactive training will provide an overview of medical device biocompatibility with a focus on entire biological evaluation process from development of a biological evaluation plan through testing and biological evaluation report according to ISO 10993-1:2025 requirements

From regulatory intent to practical execution

Risk identification, justification, and expectations

• Delving into the regulatory landscape for AI-enabled medical devices in China, including key requirements and recent updates.
• Discuss challenges and opportunities for manufacturers navigating China’s unique regulatory framework for AI-driven technologies.

• Rapid growth of software in medical devices is exposing a critical gap between Software Development (Agile, CI/CD, automated testing, modern tools) and Quality & Regulatory Affairs expectations aligned with IEC 62304, leading to misunderstandings and inefficiencies.
• Differing mindsets—developers viewing documentation as a burden and QARA teams expecting traditional, document-heavy compliance—often create redundancy, inconsistencies, and misalignment between “official” documentation and the actual software state.
• Christophe will share practical strategies to successfully bridge this gap by aligning modern development practices with regulatory compliance and integrating Quality by Design principles into everyday workflows.

Aligning scientific rationale with ISO 10993 requirements

Common pitfalls & resolution