Comment from Evalion

Evalion submits this comment on Docket FDA-2026-N-4390 in support of the AI-Enabled Optimization of Early-Phase Clinical Trials Pilot Program. This comment makes two related proposals to FDA: (1) Require evidence-based reliability for AI agents in consequential roles. Where an AI agent performs a consequential function in a trial — determining eligibility, flagging a safety signal, verifying source data — the pilot should require that its reliability be demonstrated through evaluation rather than asserted, with the rigor of that evidence scaled to the risk of the agent’s actions. We propose a concrete, inspectable evidence package: adversarial offline evaluation before live use, continuous online monitoring after deployment, automated grading calibrated against domain-expert judgment, and transparent reporting. Critically, because a reliability score is only as trustworthy as the method that produced it, the quality of the evaluation itself must be measured — a point we substantiate empirically in the attached materials. (2) Explore continuous operational oversight as a complementary capability. Building on the first proposal, the pilot is well-positioned to evaluate how AI integrated with existing trial systems (eSource, EDC, CTMS, laboratory systems) can provide continuous oversight of trial conduct — detecting protocol deviations, eligibility errors, delayed safety reporting, laboratory discrepancies, and incomplete documentation earlier than periodic monitoring allows, while preserving sponsor accountability under ICH E6(R3). The substantive comment, including detailed responses to the questions in Section II.B of the RFI, is filed as the attached document. Respectfully submitted, Miguel Andres, PhD Co-founder and Chief Technology Officer, Evalion Lugein Al Khashlok, MD Clinical Lead, Evalion