Two-axis FDA framework showing GenAI-enabled medical-device risk increasing with activity autonomy and consequences.
FDA’s proposed two-axis framework considers device activity and the consequences of relying on an incorrect output.Source: U.S. FDA/CDRH

FDA issued a discussion paper on medical software that uses generative AI. It sketches possible approaches to risk assessment, testing before launch, and monitoring afterward. It is a request for feedback, not draft or final guidance and not a policy change.

What happened

  • FDA's Digital Health Center of Excellence opened a public discussion on how to oversee medical software that generates content or actions.
  • The paper considers a two-axis risk heuristic, premarket benchmarking plus clinical confirmation, and risk-proportionate postmarket monitoring.

Why it matters

The paper identifies the questions that may shape future evidence expectations for adaptive, foundation-model-based, and agentic medical software.

Key details

Document status
Discussion paper; not draft or final guidance
Scope
Risk assessment, premarket evaluation, postmarket monitoring, foundation models, and agentic systems

What to watch

  • Whether public feedback produces a more formal proposal with testable regulatory expectations.

Event chronology

How the story developed

  1. Discussion paper issued

    FDA published the paper and opened a request for stakeholder feedback.

Evidence ledger

Sources behind this brief

  1. 01
    Primary source

    FDA discussion paper and request for feedback

    primary source · Accessed August 24, 2026