AI/ML Action Plan
FDA's AI/ML Action Plan (January 2021)
FDA published its Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device Action Plan in January 2021, outlining five key action areas to advance the regulatory framework for AI/ML-based SaMD.
The five action areas
1. Tailored regulatory framework for AI/ML-based SaMD
FDA committed to developing a risk-based framework specifically for AI/ML SaMD. The core challenge: AI/ML algorithms can adapt and change after deployment ("adaptive AI") in ways that traditional static software does not. FDA's response has been the Predetermined Change Control Plan (PCCP) framework (see PCCP guidance).
2. Good Machine Learning Practice (GMLP)
FDA (with Health Canada and MHRA) published 10 Guiding Principles for GMLP in 2021. These principles address:
- Multi-disciplinary expertise in design and development
- Clinical study design relevant to AI/ML performance
- Transparency of the algorithm and its limitations
- Dataset representativeness and management of bias
- Testing across conditions (subgroups, edge cases)
3. Patient-centric approach
FDA committed to ensuring that the perspectives of patients — including diverse populations — are incorporated into AI/ML regulatory thinking. This includes guidance on health equity considerations in AI/ML development.
4. Regulatory science research
FDA committed to investing in tools and methods to evaluate AI/ML algorithms, including:
- Methods for assessing algorithm performance across diverse populations
- Frameworks for handling algorithm drift and degradation
- Real-world performance monitoring methods
5. Real-world performance monitoring
FDA identified the need for frameworks to enable ongoing monitoring of AI/ML SaMD performance in real-world use — distinct from the pre-deployment testing framework. Real-world performance monitoring (also called postmarket surveillance or real-world evidence generation) is critical because algorithm performance may degrade over time due to data drift, population shifts, or environmental changes not captured during premarket clinical testing.
Current status (as of 2026)
| Action | Status |
|---|---|
| PCCP Guidance finalised | ✅ Complete (2024) |
| GMLP Guiding Principles published | ✅ Complete (2021) |
| Software Functions Guidance (§ 520(o)) | ✅ Finalised (2023) |
| AI/ML SaMD Action Plan discussion paper | ✅ Published (2021) |
| Real-world performance monitoring guidance | 🔄 In development |
| GMLP detailed guidance | 🔄 In development |
FDA's position on AI/ML in devices
FDA has cleared and approved numerous AI/ML-based devices — primarily in radiology, cardiology, and ophthalmology. As of early 2024, FDA has cleared or approved approximately 500+ AI/ML-enabled medical devices, with the number continuing to grow. The vast majority have been cleared via 510(k).