IMDRF publishes draft framework on best practices for AI in medical devices

The International Medical Device Regulators Forum (IMDRF) has released a draft technical document, titled ‘Technical Framework for Artificial Intelligence Life Cycle Management’, to propose a harmonized set of best practices in mitigating risks associated with the use of artificial intelligence (AI)-enabled medical devices throughout their product life cycle.

The guidance states that “the GMLP principles describe foundational best practices for the development of AI-enabled medical devices, emphasizing areas such as data quality, model transparency, performance evaluation, and the role of multidisciplinary expertise. These principles underpin each step of the AI life cycle, and this document provides relevant GMLP references to help provide a foundational understanding of applicable principles.”

At the IMDRF forum in Singapore, IMDRF announced that its management committee had agreed to publish the draft, which builds on previous guidance regarding Good Machine Learning Practice (GMLP) issued in 2025.

The draft outlines several ‘universal concepts’ applicable to the AI life cycle, including the implementation of a Quality Management System (QMS), risk management strategies, human oversight, and cybersecurity.

In terms of QMS, the guidance states that “AI-enabled medical devices benefit from implementation of scalable life cycle support processes that emphasize safety-focused risk management throughout all life cycle steps. For example, QMS requirements management captures functional specifications as well as clinical environment considerations, such as how AI outputs will be interpreted by healthcare providers or patients.”

The guidance identifies that risks related to AI-embedded medical devices include the ‘black box’ nature of some AI models, which makes it challenging to understand how and why certain outputs are produced or why certain decisions are made by the model.

The document emphasizes that human oversight—particularly from clinicians, healthcare providers, patients, and lay users—is crucial throughout the entire life cycle of AI-enabled medical devices. This oversight ensures that human and clinical expertise informs model development, validates real-world performance, and supports effective human-AI collaboration, all while prioritizing patient safety and enhancing clinical decision-making.

Comments on the draft will be accepted up to 10 June 2026.