FDA Guidance on AI Medical Devices: Predetermined Change Control Plans

One of the challenges regulators face with AI-enabled medical devices is that they are not static products. Unlike a traditional implant or diagnostic tool, the device that receives FDA approval may continue to evolve once it is on the market. Algorithms can be retrained, software can adapt, and performance can shift over time, which means that the device in patients’ hands may no longer be the same product FDA originally reviewed. This creates a regulatory paradox: how can agencies ensure safety and effectiveness when the technology itself is designed to change? 

Predetermined Change Control Plans (PCCPs) were introduced as a potential solution to that paradox. They allow manufacturers to map out in advance the kinds of modifications they expect their AI-driven devices to undergo, the protocols for implementing and validating those modifications, and the potential impacts on performance and safety. In theory, this gives regulators confidence that future changes will be controlled, while giving companies the flexibility to innovate without restarting the approval process each time.

Since the FDA issued its draft guidance on PCCPs for AI/ML-enabled device software functions in 2023, significant developments have occurred both domestically and internationally. These updates provide greater clarity on how PCCPs will be implemented and signal a move toward global alignment.

In December 2024, the FDA published its final guidance on PCCPs for AI-enabled device software functions. The guidance confirmed the PCCP framework outlined in the earlier draft and clarified expectations in several key areas. PCCPs must include three essential components:

• Description of Modifications: a detailed account of the specific changes anticipated.
• Modification Protocol: the methodology and validation steps that will be used to implement those changes safely and effectively.
• Impact Assessment: an analysis of how the modifications may affect device safety and performance.

The final guidance also expanded the scope to cover all AI-enabled devices, not just machine learning functions. It further addressed labeling, requiring transparency to users when a device has been authorized with a PCCP, and emphasized early interaction with the agency and robust risk management as manufacturers develop these plans.

In August 2025, the FDA, in collaboration with Health Canada and the UK’s Medicines and Healthcare products Regulatory Agency (MHRA), issued five guiding principles for PCCPs in machine learning-enabled medical devices. The principles underscore that PCCPs should be:

1. Focused, which means they are limited to clearly defined and verifiable modifications.
2. Risk-based, ensuring patient safety remains central.
3. Evidence-based, supported by appropriate data and validation.
4. Transparent – communicated clearly to regulators and stakeholders.
5. Lifecycle-oriented, enabling continuous innovation while maintaining oversight.

This multinational effort reflects a broader push for harmonization and consistent expectations across regulatory authorities. With the release of FDA’s final guidance and the joint international principles, PCCPs have evolved from a draft concept into a concrete regulatory framework. These developments provide manufacturers with a clearer path for implementing planned post-market modifications while safeguarding safety and effectiveness, both in the U.S. and globally.

But it is important to note that PCCPs also raise questions about products liability. If a device is allowed to change after it is already on the market, how do manufacturers keep users properly informed? How do companies evolve their warnings and instructions to avoid failure-to-warn claims? In a follow-up post, I will explore these products liability implications in more detail.