The AI in SaMD Trifecta: What FDA’s Latest Moves Mean for Your AI-Enabled Product Roadmap

PDF

Predetermined Change Control Plans, Lifecycle Expectations, and New Transparency Obligations

The FDA has been steadily laying the groundwork for a comprehensive framework around artificial intelligence (AI) in software as a medical device (SaMD). The culmination—at least for now—arrived in late 2024 and early 2025 in the form of two closely linked documents. For companies building or updating AI-enabled medical devices, these developments are not abstract policy. They directly shape how you design your models, structure your submissions, and plan your product roadmap.

This article breaks the new framework into three “legs”:

1. The Predetermined Change Control Plan (PCCP) Final Guidance
2. The Draft AI-Enabled SaMD Lifecycle Guidance
3. The Transparency Obligation that ties them together

We close with practical implications for your roadmap, especially relevant if you are working with generative AI, foundation models, or other emerging architectures.

Leg 1: The PCCP Final Guidance (December 2024)

FDA’s December 2024 final guidance on Predetermined Change Control Plans (PCCPs) is now a cornerstone of AI/SaMD regulation. It confirms a three-part PCCP structure and significantly broadens the concept’s reach.

A PCCP, submitted as part of your original marketing submission, must include these three required components:

1. Description of Modifications: a clear, bounded description of the specific changes you anticipate making after clearance
2. Modification Protocol: your methodology and validation approach for implementing those changes safely and effectively
3. Impact Assessment: an analysis of how each anticipated category of change may affect device safety and performance

Critically, the final guidance expanded scope from the draft to cover all AI-enabled devices—not just machine learning functions—and addressed labeling, requiring transparency to users when a device has been authorized with a PCCP. 

The practical upside: The FDA reviews the PCCP as part of the original marketing submission, so authorized modifications can be implemented without necessitating additional marketing submissions for each change. For companies iterating fast, this is a significant unlock, but it requires doing the upfront work of mapping your anticipated change landscape before you submit.

Leg 2: The Draft AI-Enabled SaMD Lifecycle Guidance (January 2025)

On January 7, 2025, FDA released draft guidance titled: “Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations,” which reveals FDA’s vision of what a Total Product Life Cycle (TPLC) approach to AI-enabled SaMD looks like in practice.

The guidance outlines a set of recommended content elements for marketing submissions involving AI-enabled device software functions, including:

• Model Description
• Data Lineage and Splits
• Performance Tied to Claims
• Bias Analysis and Mitigation
• Human–AI Workflow
• Monitoring and Real-World Performance Management
• PCCP (if applicable)

Collectively, these elements reinforce that the FDA is not only evaluating the model at the time of clearance, but how you will manage it responsibly over its entire lifecycle.

Labeling this a “Total Product Life Cycle” approach is not semantic. It signals that FDA expects to see not just what the product does at clearance, but how you’ll manage it over time.

The comment period for the draft guidance closed April 7, 2025. Based on historical timelines, final guidance could emerge any time now, as it typically takes 6-12 months to finalize guidance.

Leg 3: The Transparency Obligation

A common thread running through both the PCCP final guidance and the AI lifecycle draft guidance is a heightened transparency expectation toward both regulators and end users. Manufacturers must ensure that users are informed when a device operates under an authorized PCCP. This is not merely a box-checking exercise. It has practical implications for labeling, training materials, user education, and communication of update histories. The regulatory rationale is straightforward: when models can change after clearance, users need a reasonable understanding of that dynamic to use the device safely and appropriately.

The draft AI lifecycle guidance also emphasizes transparency with FDA around:

• Data lineage: where data came from, how it was processed, and how it maps to real-world use
• Performance tied to claims: explicit, evidence-backed mapping of metrics to specific labeled claims
• Performance across demographic groups: requiring sponsors to assess whether performance is consistent across relevant demographic subgroups and to address disparities

In practice, this can mean several things, including planning your data strategy early so you can credibly evaluate subgroup performance, incorporating fairness/bias analysis into your standard validation package—not as an afterthought, and designing monitoring mechanisms to detect how a change might affect different populations differently over time.

What This Means for Your Product Roadmap

If there is any realistic chance that your AI model will change after clearance—whether through new training data, architecture tweaks, threshold adjustments, or deployment to new clinical contexts—you should assume a PCCP will be expected and plan accordingly.

The strategic advice: planning should start well before submission, with a strategic tool we often recommend to clients: the “Model Evolution Roadmap”. When you create the roadmap early, even if updates aren’t happening yet, it becomes a companion to training documentation and helps ensure you’re ready when real-world usage triggers update needs. 

Key Watch-out: Comments on the January 2025 draft guidance underscored concerns about emerging technologies, including:

• Generative AI (e.g., text, image, or report generation)
• Foundation models and large language models (LLMs)
• Multimodal AI systems that combine, for example, text, imaging, and sensor data

The FDA has acknowledged these concerns and signaled that its current recommendations may need to evolve to fully address them.

The takeaway: while the current framework provides a structure you can use today, regulation in this area is still evolving, and product teams should build flexibility into their regulatory and technical strategies.

Action Items for AI-Enabled SaMD Developers

1. Decide Now Whether You Need a PCCP
   • If post-market model changes are likely, begin drafting your Description of Modifications, Modification Protocol, and Impact Assessment.
2. Develop a Model Evolution Roadmap
   • Align engineering, regulatory, and clinical leadership on anticipated model evolution over the next several years.
3. Strengthen Data and Validation Infrastructure
   • Ensure you can document data lineage, run robust validation (including subgroup analyses), and generate submission-ready evidence tied to labeling claims.
4. Design for Transparency and User Communication
   • Plan labeling, user education, and update communication strategies to meet FDA’s transparency expectations, particularly where PCCPs are involved.
5. Monitor Regulatory Developments—Especially for Generative and Foundation Models
   • Track finalization of the AI lifecycle guidance and any additional FDA communications on generative AI, LLMs, and multimodal systems.

By treating FDA’s evolving AI/SaMD framework as a design input—rather than a late-stage constraint—companies can maintain agility, accelerate iteration, and reduce regulatory friction, all while demonstrating a defensible commitment to safety, effectiveness, and equity.