The FDA's Real-Time Clinical Trials (RTCT) initiative moves safety data from periodic batches toward continuous flow. At the same time, AI has spread through pharmacovigilance faster than the governance around it. Read together, the regulatory signal is consistent even where it is not yet spelled out: a flag is not a decision, and an AI output is not a record of one.

The expectation underneath the language is that a qualified physician adjudicates benefit-risk, and that the review leaves an attributable, timestamped, inspection-ready trail — who reviewed, when, under what criteria, with what outcome. As detection accelerates, the human-oversight obligation does not shrink; it becomes more visible, and more inspectable.

For a sponsor, the practical translation is to be able to answer one question on demand: show me the documented physician review for these AI-flagged signals. Programs that can produce that answer in the meeting are ready for where RTCT is heading. Programs that have to reconstruct it are exposed.