Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products Guidance for Industry and Other Interested Parties; Draft Guidance

The commenter is a regulatory medical writing lead for a pharmaceutical sponsor seeking specific clarification on the FDA's stance on using AI for authoring various regulatory documents. They are asking for guidance on whether AI-generated content is permissible when human-oversight is maintained and how such usage should be documented.

A pharmaceutical or biotechnology sponsor (implied by the use of "sponsors" and "industry" context) supports the draft guidance for using AI in regulatory decision-making. They suggest that the FDA provide further clarification on how to document human accountability and adequacy determinations when AI models are integrated as automated, embedded components of regulatory workflows.

The commenter provides specific technical suggestions to improve the draft guidance on AI in regulatory decision-making. They request more examples of "operational efficiency" exclusions and seek clarification on data traceability, validation expectations for federated learning, and validation approaches for unsupervised models.

The commenter suggests that the FDA clarify the scope of the guidance to include AI systems used to generate or transform data submitted to the Agency, even if their primary purpose is operational. They argue this will improve alignment and prevent misclassification of AI systems.

Dr. Ajay Babu Pazhayattil suggests broadening the scope of the guidance to include AI uses for supply chain management and document writing, rather than limiting it to safety, effectiveness, and quality. He also proposes specific re-wording regarding manufacturing controls to better reflect modern industry practices and input parameter-based controls.

Certara, a pharmaceutical services company, supports the draft guidance but requests more specific details and clarity. They specifically ask for a clearer risk matrix, more complex examples, precise definitions for "training" versus "tuning," and recommendations for specific meeting types (like Type D meetings) to discuss AI development with the FDA.

eSTARHelper LLC, a software developer, proposes a "dueling banjo" strategy that combines AI/ML-driven insights with human-led full-text regulatory searches to support regulatory decision-making. They argue that this human-in-the-loop approach, utilizing their proprietary SmartSearch+ and an FDA Copilot, ensures accuracy, mitigates AI hallucinations, and improves the efficiency of preparing regulatory submissions like Quality Overall Summaries.

The commenter suggests including specific language regarding Large Language Models (LLMs) in the draft guidance to highlight their role in generating reports and protocols. They advocate for distinguishing between public LLMs, which they view as risky, and internal, site-specific LLMs that can support junior engineers and improve operational efficiency.