Comment from Rajat Rawal

I strongly support the FDA’s initiative to evaluate how artificial intelligence can responsibly improve the efficiency and quality of early-phase clinical trials while maintaining rigorous scientific and regulatory standards. The emphasis on trustworthy AI principles aligned with the NIST AI Risk Management Framework provides an appropriate foundation for this effort. I would like to offer my recommendations for consideration. First, AI systems participating in the pilot should be evaluated as continuously managed systems rather than static software tools. Model performance can change over time as clinical practices evolve, new data become available, or patient populations differ from those used during development. Accordingly, the pilot should encourage sponsors to establish ongoing monitoring for model performance, bias, calibration, and model drift throughout the lifecycle of the clinical trial instead of relying solely on pre-deployment validation. Second, reproducibility and auditability should receive substantial attention. Every AI-assisted recommendation that contributes to recruitment, dose selection, safety monitoring, or other trial decisions should be traceable to the specific model version, input data, and validation evidence used at that point in time. This level of documentation will strengthen regulatory confidence and facilitate independent review when important clinical decisions are evaluated. Third, pilot success should be measured using both operational and clinical outcomes. In addition to reductions in enrollment time or improvements in predictive accuracy, evaluation should include measures such as earlier identification of safety signals, improved data completeness, fewer protocol deviations, consistency of AI-supported recommendations across trial sites, and the level of agreement between AI-supported and expert clinical decisions. Finally, I encourage FDA to include organizations with varying levels of AI maturity and technical resources. Demonstrating that appropriate governance practices can be implemented by both large and smaller organizations will improve the scalability and long-term value of the pilot while encouraging broader adoption of responsible AI practices. Artificial intelligence has significant potential to improve early-phase clinical research, but its value will ultimately depend on consistent governance, transparency, and ongoing oversight. A pilot that evaluates not only model performance but also operational reliability and lifecycle management will provide valuable evidence for future regulatory guidance and help establish practical expectations for trustworthy AI in clinical research.