FDA’s Quality Management System Regulation (QMSR) took effect on February 2, 2026. It replaced the old Quality System Regulation in 21 CFR Part 820 by incorporating ISO 13485:2016 by reference. Three points frame everything else:

Five Checks for AI-Enabled Device Teams

  1. Treat models, datasets and ML platforms as suppliers. ISO 13485 clause 7.4 expects criteria for selecting, monitoring and re-evaluating suppliers. If you use a third-party model, a labelling vendor or a cloud ML service, can you show how you qualified and monitor it?
  2. Validate the software that runs your quality system and production. ISO 13485 requires risk-proportionate validation of software used in the QMS, in production and service provision, and in monitoring and measurement. AI-based tools used there are included.
  3. Run AI-specific failure modes through risk management. Design and development under clause 7.3 must include risk management. For an AI product that means model drift, performance gaps in subgroups, and unexpected inputs, not only hardware and software faults.
  4. Make internal audits cover the AI lifecycle. An audit programme that checks document control and CAPA but never samples data management, model validation or change control will leave gaps that are now visible to an inspector.
  5. Trace each release end to end. Training data, model version, validation results and release approval should link together in the design records. If the model changes after release, your change control and any predetermined change control plan should show who approved what, and why. See our earlier piece on what a PCCP actually requires.
// THE POINT

None of these are new ideas. What changed is that the standard behind them is now the law an inspector works from, and AI-enabled products stretch every one of them.

This is general information, not legal or regulatory advice. Sources: FDA, Quality Management System Regulation (QMSR); IntuitionLabs, FDA QMSR and ISO 13485: Key Changes Effective 2026.