Glossary Updates16 new terms added to the glossaries · November 12, 2026, 21:23 CET
AI TechDocKnowledge

Medical-device documentation

Software and AI lifecycle documentation

Also known as: Medical device software documentation · AI-enabled device documentation · IEC 62304 documentation

Medical device software is documented along its lifecycle — planning, requirements, architecture, implementation, verification, release, maintenance and problem resolution — under IEC 62304, within the quality management system and the risk management of the device. Software that uses artificial intelligence adds records on data, models, performance and change. In the EU, an AI system that is a medical device or a safety component of one and requires notified-body conformity assessment is a high-risk AI system under the AI Act; its AI Act documentation is part of the single set of technical documentation drawn up with the MDR or IVDR file (Article 11(2)).

By knowledge.aitechdoc.world · Last reviewed

Definition

IEC 62304:2006 with Amendment 1:2015 defines lifecycle processes for medical device software and classifies software systems into safety classes A, B and C according to the possible harm. Software of unknown provenance (SOUP) is software already developed and generally available that was not developed for the purpose of being incorporated into the device, or whose development records are not adequate. The International Medical Device Regulators Forum (IMDRF) defines software as a medical device (SaMD) as software intended for one or more medical purposes that performs these purposes without being part of a hardware medical device. The US FDA uses the term AI-enabled device software function (AI-DSF) for a device software function that implements one or more AI models.

Regulatory framework

  • EU. Software is classified under Rule 11 of Annex VIII MDR (and the IVDR rules). Under Article 6(1) of the AI Act (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744), an AI system is high-risk when it is a product, or a safety component of a product, covered by legislation in Annex I Section A — which includes the MDR and IVDR — and that product requires third-party conformity assessment. The use of AI alone does not make a device high-risk under the AI Act; a class I device assessed by the manufacturer alone does not meet the condition. For such devices the high-risk requirements apply from August 2, 2028. Article 11(2) requires a single set of technical documentation containing the information of the AI Act's Annex IV and of the sectoral legislation, and the quality management elements can be integrated into the MDR quality management system. MDCG 2025-6 (AIB 2025-1) explains the interplay.
  • USA. Software functions may be device functions, device software functions excluded from the device definition by section 520(o) FD&C Act, or subject to enforcement discretion. The FDA's guidance on clinical decision support software, revised in January 2026, explains the criteria for non-device CDS. A predetermined change control plan (PCCP) under section 515C FD&C Act lets planned modifications, such as retraining within a defined protocol, be authorized with the marketing submission; the FDA's final guidance on PCCPs for AI-enabled device software functions describes its content.

Components

  1. Software development plan, requirements and architecture, with the software safety class and the SOUP list.
  2. Verification records: unit, integration and system testing, anomalies and their evaluation.
  3. Release and configuration records: the version of each software item and the configuration placed on the market.
  4. Maintenance and problem resolution, including the evaluation of field feedback and SOUP updates.
  5. For AI: intended purpose of the model, data management and provenance of training, tuning and test data, representativeness for the intended population, performance metrics and limits, human oversight measures, transparency information for users, logging and post-market performance monitoring.
  6. Change documentation: for the EU, the assessment of whether a change is significant under the MDR or IVDR and whether it was predetermined in the technical documentation under the AI Act; for the USA, the PCCP or a new submission.

Function and purpose

The records show which software, model and data produced which released device configuration and on what evidence. Model version, software version and device version are distinct: a retrained model can produce a new software build without changing the device's identification, or it can change performance enough to affect clinical evidence and labeling. Documentation that keeps these versions apart allows each statement in the file — a performance claim, a limitation, a warning — to be tied to the configuration for which it holds.

Limitations

No documentation format makes a learning or changing model safe. Changes outside a predetermined plan or the assessed scope remain subject to the applicable change procedures, and the use of AI does not reduce MDR or IVDR obligations, which apply regardless of the AI Act.

Conclusion

For software and AI, the technical documentation is a record of versions and their evidence. Its central question is which statement holds for which configuration, and who released it.

Further reading

Cite this article

knowledge.aitechdoc.world. “Software and AI lifecycle documentation.” Encyclopedia, AI TechDoc Knowledge. Last reviewed November 10, 2026. https://knowledge.aitechdoc.world/encyclopedia/medical-device-documentation/software-and-ai-lifecycle-documentation