FDA CDRH provides regulatory requirements and guidance for medical devices incorporating artificial intelligence (AI) technology. AI-enabled medical devices offer potential benefits including improved diagnostic accuracy and enhanced clinical efficiency; however, algorithm transparency, validation, and continuous monitoring are critical to safe and effective performance. FDA expects manufacturers to conduct machine learning model validation and implement comprehensive risk management strategies. The guidance addresses the unique challenges posed by adaptive algorithms and evolving AI systems in the medical device lifecycle, emphasizing the importance of rigorous testing methodologies and post-market surveillance protocols. Manufacturers must demonstrate that AI systems maintain performance across diverse patient populations and clinical conditions. The document also discusses design controls, software development practices, and documentation requirements specific to AI/ML-based medical devices under 21 CFR Part 820 and FDA premarket submission pathways.

FDA has updated its searchable database of medical devices incorporating digital health technology that have received U.S. marketing authorization. The updated lists include AI/ML-enabled Software as a Medical Device (SaMD) products. This reference provides manufacturers and regulators with current visibility into approved digital health and algorithmic medical devices, supporting transparency in regulatory decision-making and market surveillance activities.

This FDA draft guidance (2025) provides lifecycle management and premarket submission recommendations for software incorporating artificial intelligence and machine learning (AI/ML) technologies. The document addresses critical AI-specific considerations including training data management, algorithm performance monitoring, software modification processes, and predetermined change control planning. Manufacturers should establish robust procedures for documenting training and validation datasets, monitoring real-world performance against predetermined performance specifications, and implementing planned modifications without requiring submission of new premarket applications when modifications fall within pre-approved change control plans. The guidance specifically addresses adaptive algorithms that modify behavior based on accumulated clinical experience, establishing frameworks for distinguishing routine algorithm refinement from material modifications requiring FDA notification. Post-market performance evaluation plans should establish metrics for ongoing algorithm performance assessment across diverse patient populations and clinical settings. The document remains under comment collection, with FDA inviting stakeholder input on implementation feasibility and technical approaches. Manufacturers of AI/ML-enabled medical devices should actively monitor guidance finalization and incorporate recommendations into development strategies to facilitate efficient regulatory approval pathways.