FDA AI Medical Device Regulations in 2025 — What's Changed and What's Coming

FDA's regulatory framework for AI medical devices has evolved more in the past three years than in the preceding decade. The combination of the PCCP final guidance, the AI-Enabled Device Software Functions draft, and an uptick in enforcement has changed what FDA expects from AI device submissions — and what it will take to stay compliant post-market.

This is a current-state overview of where FDA stands on AI devices, based on published guidance documents, enforcement actions, and clearance patterns.

The Foundational Framework: Where Things Stand

FDA regulates AI medical devices primarily under the existing 510(k) and De Novo frameworks, with AI-specific special controls layered on top through De Novo grants. There is no standalone AI device statute — FDA applies existing medical device law to AI systems and has issued guidance to clarify how.

The core documents that govern AI device regulation today:

• Predetermined Change Control Plans — Final Guidance (January 2025): Establishes requirements for PCCP submissions, allowing sponsors to pre-specify algorithm changes that won't require a new 510(k)
• AI-Enabled Device Software Functions — Draft Guidance (June 2024): Clarifies when AI outputs constitute device functions requiring clearance versus administrative or operational functions that don't
• Good Machine Learning Practice (GMLP) Guiding Principles (April 2023): 10 principles for AI/ML lifecycle management, co-issued with Health Canada and MHRA
• Action Plan for AI/ML-Based Software as a Medical Device (January 2021): The foundational document outlining FDA's five-part approach to AI device oversight

The PCCP Final Guidance: What It Actually Requires

The January 2025 PCCP final guidance is the most operationally significant document FDA has published for AI devices. It gives manufacturers a mechanism to pre-authorize algorithm updates without requiring a new 510(k) for each change — but the bar is high.

A compliant PCCP must specify: the types of modifications anticipated, the performance goals that must be maintained, the methods for assessing whether performance goals are met, and the controls that ensure the device continues to perform safely. Vague PCCPs that don't specify actual performance thresholds are being rejected by FDA reviewers.

Practical implication: if you're planning to retrain your model on new data or expand to new patient populations post-clearance, you need a PCCP. Submitting without one and then modifying your algorithm is the fastest path to an enforcement action.

The Draft Software Functions Guidance: A Key Clarification

The June 2024 draft guidance on AI-Enabled Device Software Functions addresses a question that has caused significant confusion: when does an AI output constitute a "device function" requiring FDA clearance?

FDA's position: an AI function is a device function — and requires clearance — when it is intended to diagnose, treat, cure, mitigate, or prevent a disease or condition, and the AI output drives or meaningfully informs clinical decision-making. Administrative AI (scheduling, billing, documentation assistance) is not a device function. Triage AI that flags patients for clinician review occupies a gray zone that depends on how the output is intended to be used.

Enforcement Trends: What FDA Is Actually Acting On

Enforcement actions against AI devices have increased significantly since 2023. The most common violations:

• Marketing AI diagnostic software without 510(k) clearance — the most common warning letter basis
• Algorithm updates deployed post-clearance without 510(k) or PCCP coverage — increasingly scrutinized
• Claims that exceed cleared indications — AI devices being marketed for indications beyond what FDA cleared
• Failure to report malfunctions or adverse events involving AI outputs — MDR compliance for AI devices

The enforcement pattern suggests FDA is prioritizing AI devices that make diagnostic claims, operate autonomously, or have been updated post-clearance without appropriate regulatory coverage.

What to Watch in 2026

Several developments are on the horizon that will affect AI device regulation:

The GMLP principles are expected to evolve from voluntary guidelines into more prescriptive requirements, particularly around training data diversity and post-market performance monitoring.

FDA has signaled interest in establishing real-world performance monitoring requirements for AI devices, particularly high-risk autonomous diagnostic AI. Expect draft guidance on post-market surveillance for AI devices.

International harmonization is accelerating. The EU AI Act, which came into force in 2024, applies to AI medical devices and imposes requirements that in some areas exceed FDA's current expectations. Companies planning international commercialization need to monitor both frameworks.

Practical Checklist for 2025 AI Device Submissions

• Identify your product code and the De Novo that established it — understand the special controls
• Prepare a PCCP if you anticipate any post-clearance algorithm modifications
• Ensure your training and test datasets meet emerging diversity expectations
• Define your intended use precisely — avoid claims that could be interpreted as expanding beyond your cleared indication
• Establish post-market surveillance procedures before clearance, not after