Comment from Anonymous

The draft guidance provides a clear and valuable framework for evaluating the credibility of AI model outputs intended to support regulatory decision making. In particular, the emphasis on defining the Question of Interest, the Context of Use (COU), and the assessment of model influence and decision consequence is highly constructive for sponsors adopting advanced AI enabled methodologies. As AI capabilities mature, an increasing number of regulatory support workflows may involve highly automated, end to end integration of AI generated outputs, rather than isolated or tool level use. In such scenarios, AI models may generate complete analytical artefacts or regulatory support materials that are operationally embedded into time compressed development or submission processes, with minimal manual friction. In these system embedded contexts, an important interpretive question arises regarding how and where explicit adequacy determination should be documented, particularly when AI outputs are functionally treated as default inputs to downstream regulatory activities. While the draft guidance appropriately states that model influence and decision consequence should drive the level of credibility assessment, additional clarification may be helpful on the following points: •Whether FDA expects explicit, human documented adequacy determinations when AI models are integrated as default components of regulatory workflows, rather than used as discrete analytical tools. •How sponsors should distinguish between AI generated recommendations and human acceptance of those recommendations, especially when system design implicitly reduces traditional review touchpoints. •Whether the Agency views the absence of a clearly identifiable human adequacy determination event as a potential risk indicator in high influence, high consequence AI use cases. Clarifying these expectations could assist sponsors in designing automation enabled workflows that preserve human accountability and transparency, while still realizing the efficiency gains offered by advanced AI systems. Such clarification would also help ensure consistent interpretation of the guidance across sponsors employing differing levels of workflow automation. The draft guidance has an important opportunity to illustrate how human accountability for adequacy determination can be made visible and auditable, even as AI models become increasingly embedded within regulatory decision support systems. Additional examples or language addressing this boundary would further strengthen the guidance and provide practical direction to industry. Thank you for the opportunity to comment on this draft guidance. The framework described has the potential to meaningfully advance responsible and scalable use of AI in regulatory decision making, and further clarification on system embedded AI workflows would enhance its long term applicability.