Can you update your AI medical device without a new FDA submission?

Plan the update strategy before the first submission

AI product teams often expect to improve their model after launch. A regulated medical-device roadmap needs to account for that ambition before release. FDA’s final guidance for Predetermined Change Control Plans (PCCPs) for AI-enabled device software functions was issued on August 18, 2025. It offers a framework for describing planned modifications and how they will be evaluated.

For a founder, this raises an early product question: which improvements can the team define and validate in advance, and which are still exploratory? Writing that boundary down is more useful than promising investors that the model will simply keep learning.

Sources: FDA: Final AI-enabled device software PCCP guidance (August 18, 2025; PDF)

What an authorized PCCP can cover

A PCCP is reviewed as part of an FDA marketing submission, including applicable 510(k), De Novo or PMA pathways. When FDA authorizes the plan, changes within its scope that are implemented according to the authorized plan may proceed without a new marketing submission for each change.

A plan written internally is not an authorized PCCP. An update outside its bounds still needs an assessment under the applicable change requirements. The FDA framework does not turn a cleared medical device into an unrestricted software service, and it does not establish authorization in other countries.

Sources: FDA: Predetermined Change Control Plans overview

The three parts your team needs to explain

The FDA guidance describes three main components. The Description of Modifications identifies the planned changes. The Modification Protocol explains the methods for developing, validating and implementing those changes. The Impact Assessment evaluates the benefits and risks, including the combined effect of the modifications.

For planning, translate those headings into engineering questions: What exactly changes? Which evidence will demonstrate acceptable performance? Who checks the result and approves deployment? What will users need to know? A vague commitment to “improve accuracy” does not tell a reviewer where the update boundary sits.

Sources: FDA: Final AI-enabled device software PCCP guidance (August 18, 2025; PDF)

Treat validation and transparency as product requirements

The FDA, Health Canada and the UK MHRA’s joint guiding principles emphasize focused, risk-based and evidence-based plans, transparency and a total-product-lifecycle approach. These principles are useful when deciding whether a proposed update program is sufficiently bounded and supported.

Our practical recommendation is to connect the regulatory plan to the release workflow. Keep the model version, datasets, acceptance criteria, test results and deployment decision together. Make someone responsible for checking performance in the intended population, documenting failures and deciding when a proposed release must stop for further review.

For example, compare a specified performance adjustment within a defined intended use with a proposal to add a new clinical indication. Both might be attractive commercially, but the second should never be assumed to fit the first plan. The actual authorized wording and product-specific requirements control the assessment.

Sources: FDA, Health Canada and MHRA: PCCP guiding principles

Five questions before the next model release

Use these questions to organize a discussion with your regulatory and quality leads. They are a planning aid, not a substitute for reviewing the authorized plan and the applicable FDA requirements.

• Do we have an FDA-authorized PCCP, and which version applies to this device?
• Is this exact modification within the authorized scope?
• Have all required evaluation methods and acceptance criteria been satisfied?
• Are implementation, user communication and monitoring steps ready?
• If the change falls outside the plan, have we assessed whether a new submission is needed?