How many AI/ML devices has the FDA authorized?

Over 900 as of 2026, with radiology the largest category. The number continues to grow.

What is the difference between 510(k) and De Novo?

510(k) requires substantial equivalence to an existing predicate. De Novo is for novel low-to-moderate risk devices without a predicate, creating a new classification.

Can AI medical devices be updated after FDA clearance?

Yes, the PCCP framework allows describing planned modifications in the original submission and implementing them without new applications, if changes follow the approved protocol.

What are GMLP principles?

Ten guiding principles by FDA, Health Canada, and UK MHRA covering multi-disciplinary development, representative data, independent datasets, and real-world performance monitoring.

What post-market requirements apply?

Manufacturers must monitor real-world performance, track model drift, watch for subpopulation bias, report adverse events, and maintain documentation supporting PCCP modifications.

Quick answer

The FDA regulates AI/ML-based SaMD through existing premarket pathways (510(k), De Novo, PMA) supplemented by the predetermined change control plan framework for iterative algorithm updates.

Updated June 2026 · MmowW AI Compliance

FDA AI/ML SaMD Guidance: Compliance Framework for Software as a Medical Device (2026)

FDA Regulation of AI/ML-Based Software as a Medical Device

The FDA has developed a tailored regulatory approach for AI/ML-based Software as a Medical Device (SaMD). This framework balances rigorous safety oversight with recognition that AI/ML systems improve through iterative learning. As of 2026, the FDA has authorized over 900 AI/ML-enabled medical devices spanning radiology, cardiology, ophthalmology, and numerous other specialties.

Classification and Premarket Pathways

AI/ML SaMD follows the same classification system as other medical devices, categorized into Class I, II, or III based on risk.

Pathway	Typical Class	When Used	Review Type
510(k)	Class II	Substantially equivalent to predicate	Premarket notification
De Novo	Class I or II	Novel, low-moderate risk, no predicate	Risk-based classification
PMA	Class III	High-risk devices	Full premarket approval

Submissions for AI/ML SaMD should address algorithm architecture and training methodology, datasets used for training and validation, performance metrics, and failure mode characterization.

The Predetermined Change Control Plan

The PCCP framework represents the FDA's most significant adaptation for AI/ML devices. Traditional regulation assumes a fixed product. AI/ML systems improve over time. A PCCP allows manufacturers to describe planned modifications in their original submission and receive authorization to implement changes without new premarket applications.

PCCP Components

Description of planned modifications (improved performance, expanded populations)
Modification protocol describing development, validation, and implementation
Impact assessment describing safety and effectiveness evaluation

Good Machine Learning Practice

The FDA, Health Canada, and UK MHRA jointly published Good Machine Learning Practice (GMLP) guiding principles covering the full AI/ML device lifecycle. Key principles include multi-disciplinary expertise, representative datasets, independent test sets, and robust real-world monitoring.

Real-World Performance Monitoring

Manufacturers must monitor post-market performance to identify degradation and support PCCP modifications. Systems should collect real-world performance data, track model drift, monitor for bias across subpopulations, and respond to safety signals.

Transparency and Labeling

Device labeling should communicate intended use, the AI's role in clinical workflow, performance characteristics, training data description, and known limitations. This enables informed decisions about integrating AI tools into clinical practice.

Practical Steps

Engage the FDA early through Pre-Submission programs
Design your QMS to address AI-specific requirements from the outset
Document data management practices including provenance and representativeness
Develop a PCCP strategy before your initial submission
Implement GMLP principles throughout development
Plan for post-market performance monitoring and reporting

Check your AI compliance readiness — free.

Take the Readiness Check 3 minutes · 10 questions · no signup required

This article is for informational purposes only and does not constitute legal advice. Regulatory requirements change frequently — verify current rules with official sources. Built by Sawai Gyoseishoshi Office, Hiroshima, Japan.