The U.S. Food and Drug Administration has released updated guidance for AI-enabled medical devices that reshapes how manufacturers must approach the development, validation, and lifecycle management of artificial intelligence and machine learning-enabled medical devices. Building on several years of draft frameworks, the FDA’s 2026 regulatory posture consolidates clearer expectations around transparency, real-world performance monitoring, and predetermined change control plans (PCCPs).
It is important to note that not all elements of this framework are finalized. The August 2025 final PCCP guidance is fully in effect. The January 2025 draft guidance on AI-enabled device software functions and lifecycle management remains in draft form, pending finalization. Manufacturers should treat the draft guidance as a strong signal of regulatory direction, while recognizing that certain specifics may evolve before the final version is issued.
For manufacturers already marketing AI-enabled devices or preparing submissions, these developments introduce specific documentation and post-market requirements that will affect both regulatory strategy and operational workflows. The guidance applies broadly to software as a medical device (SaMD) that incorporates AI or machine learning functionality, though higher-risk devices face more stringent expectations.
What the Guidance Covers
The FDA’s current AI regulatory framework for medical devices draws from three interconnected streams: the August 2025 final PCCP guidance, the January 2025 draft guidance on AI-enabled device software functions and lifecycle management, and the June 2024 transparency guiding principles for machine learning-enabled devices. Together, these documents cover the total product lifecycle (TPLC) approach for AI/ML-based SaMD, PCCP requirements, and transparency and bias mitigation obligations.
One of the most consequential elements is the finalized PCCP framework. Under the August 2025 final guidance, manufacturers submitting premarket applications for AI/ML devices must include a detailed predetermined change control plan that specifies the types of modifications the algorithm may undergo after clearance or approval, the methodology for validating those modifications, and the performance boundaries within which changes may occur without requiring a new submission. This codifies what had previously been a loosely defined expectation into a structured, auditable requirement.
Transparency and Labelling Requirements
The draft lifecycle management guidance introduces more prescriptive transparency obligations. Manufacturers are expected to include specific information describing the data used to train and validate the AI model, including the demographic composition of training datasets. The FDA has framed this as a measure to address algorithmic bias and to give clinicians meaningful context for interpreting AI-generated outputs.
Labeling must also clearly communicate the intended use conditions under which the device has been validated, as well as any known limitations of the algorithm’s performance across different patient populations or clinical settings. These expectations go beyond what many manufacturers have historically included in their labeling, and companies should begin assessing their current practices against this direction now, even before the draft guidance is finalized.
Real-World Performance Monitoring
Another significant addition in the draft guidance is the formalization of post-market real-world performance monitoring expectations. Manufacturers of AI/ML-enabled devices are expected to establish ongoing monitoring programs that track device performance using real-world data. This includes monitoring for model drift, which occurs when an algorithm’s accuracy degrades over time as patient populations or clinical practices change.
The draft guidance does not mandate a specific monitoring methodology but identifies several acceptable approaches, including the use of electronic health record data, device output logging, and structured feedback from clinical users. Manufacturers are expected to document their monitoring approach in their quality management system and to demonstrate during post-market surveillance reviews that they are actively tracking algorithmic performance.
Predetermined Change Control Plans in Detail
The finalized PCCP framework deserves particular attention because it directly affects how manufacturers plan their product development roadmaps. Under the August 2025 final guidance, a PCCP must define three components with specificity. First, the description of modifications outlines the types of changes the manufacturer anticipates making, such as retraining the model on new data or adjusting decision thresholds.
Second, the modification protocol details the verification and validation activities that will be performed before implementing any change. Third, the impact assessment describes how the manufacturer will evaluate whether a given change falls within the scope of the approved PCCP or requires a new regulatory submission.
This framework provides a pathway for iterative algorithm improvement without the burden of filing a new 510(k) or premarket approval supplement for every update. However, it places the responsibility on manufacturers to define their change boundaries with precision at the time of initial submission. Vague or overly broad PCCPs are likely to draw additional scrutiny from FDA reviewers.
What This Means for Your Business
For companies developing or marketing AI-enabled medical devices, the practical impact spans multiple functions. Regulatory affairs teams will need to revise submission strategies to incorporate detailed PCCPs in close collaboration with data science and engineering teams. Quality assurance teams will need to update quality management systems to include real-world performance monitoring protocols, and labeling teams should prepare for more detailed transparency disclosures.
Companies with legacy AI/ML devices already on the market should begin assessing their current labeling against the draft guidance now, even before finalization. Firms operating in both U.S. and Canadian markets should also note that the QMSR came into effect February 2, 2026, aligning U.S. quality system requirements with ISO 13485:2016, and that Health Canada continues to align with international AI/ML frameworks through the IMDRF, meaning FDA compliance investments may carry parallel benefits for Canadian filings.
Quality Smart Solutions recommends treating this regulatory direction as an operational planning milestone, not a distant concern. Compliance preparation should begin well before submission timelines.
Frequently Asked Questions
Does the current guidance apply to AI devices that already have FDA clearance?
The finalized PCCP guidance applies to new marketing submissions. The transparency and labeling expectations outlined in the draft lifecycle management guidance are expected to apply to legacy devices as well, though the FDA has indicated a phased compliance timeline for already-cleared devices. The specific deadlines will vary depending on device risk classification, and manufacturers should monitor for the finalized version of the draft guidance, which will confirm the compliance schedule.
How specific does a predetermined change control plan need to be at the time of submission?
The August 2025 final PCCP guidance makes clear that overly broad or vaguely defined change descriptions are unlikely to be accepted. A compliant PCCP must detail the specific categories of anticipated modifications, the validation methodology for each, and the quantitative performance boundaries that define whether a change remains within scope. Manufacturers should engage their data science and regulatory teams early in development to define these parameters with enough precision to satisfy FDA reviewers while retaining room for meaningful algorithm improvement.
Will Health Canada adopt similar requirements for AI-enabled medical devices?
Health Canada has been actively participating in the IMDRF and has co-published guiding principles on PCCPs for machine learning-enabled devices alongside the FDA and the UK’s Medicines and Healthcare products Regulatory Agency (MHRA). Health Canada has also published its own guidance on transparency for machine learning-enabled devices. Manufacturers selling into both markets should design their compliance programs to satisfy both jurisdictions from the outset, as the regulatory frameworks continue to converge.
Key Takeaways
- The FDA’s August 2025 final PCCP guidance is in effect, requiring detailed change control plans in new AI/ML device marketing submissions.
- The January 2025 draft guidance on AI-enabled device lifecycle management remains in draft form; manufacturers should treat it as a strong indicator of regulatory direction.
- Predetermined change control plans must be specific at the time of initial submission, defining modification types, validation protocols, and performance boundaries.
- Transparency and labeling expectations call for disclosure of training data demographics, intended use conditions, and known algorithmic limitations.
- Real-world performance monitoring programs are an explicit expectation in the draft guidance, with manufacturers expected to track model drift and document their approach within their quality management system.
- The QMSR came into effect February 2, 2026, aligning U.S. quality system requirements with ISO 13485:2016.
- Health Canada is actively aligning with international AI/ML device frameworks; manufacturers in both markets should plan for converging requirements.
Looking Ahead
The FDA’s evolving AI medical device framework marks a substantive shift in how regulators expect manufacturers to manage the full lifecycle of AI-enabled products. Companies that align their regulatory, quality, and data science functions now will be in a stronger position to meet submission and post-market requirements without delays.
For manufacturers seeking support in preparing PCCP documentation, updating quality management systems, or aligning U.S. and Canadian regulatory strategies, contact Quality Smart Solutions to speak with a regulatory specialist.
