SaMD manufacturers with AI/ML functionality now answer to two evolving regulatory frames — FDA's AI/ML SaMD guidance and EU MDR (with Annex 22-style expectations layering in). They differ in emphasis, not in destination. A single working assurance stack can meet both. Here is how they compare.
Where FDA and EU MDR overlap
- Model governance — what the model is for, what it is not for.
- Evaluation evidence — a benchmark set that gates every release.
- Safeguards on outputs — the model cannot emit unbounded content.
- Human-in-the-loop — enforced by the product, not by policy.
- Drift monitoring with a defined escalation path.
- Documented fallback when the model is unavailable.
Where FDA emphasises differently
FDA's AI/ML SaMD action plan foregrounds pre-determined change control plans (PCCPs) — a way to specify in advance what model changes are anticipated and how they will be managed. If your model is expected to learn or update, plan for a PCCP conversation.
Where EU MDR emphasises differently
EU MDR foregrounds clinical evidence and post-market surveillance. The clinical evidence dossier — including CER — is the artefact regulators walk end-to-end. Traceable claims matter more here than in the FDA conversation.
The one working assurance stack that meets both
- Published governance (satisfies both).
- Frozen, versioned benchmark set (satisfies both).
- Output safeguards (satisfies both).
- HITL enforced by the tool (satisfies both).
- Drift monitor with escalation (satisfies both).
- Documented fallback (satisfies both).
- Add: PCCP for FDA. Add: source-traceable clinical evidence via TraceDraft for MDR.
Delivered as a service
Our AI validation services team stands the stack up. See our SaMD case study.
Where to go next
Explore GxP Copilot for AI-native validation, TraceDraft for source-traceable clinical documentation, or book a demo to see either on your own data.
