Scientific oversight combines evaluating unfamiliar AI-assisted claims and methods with people's ability to question, investigate and redirect the work. Approval authority alone is insufficient if reviewers cannot assess the evidence.

Operational scalability concerns whether adequate scrutiny remains feasible as outputs multiply or become harder to assess. Asking this question does not imply that a universal oversight bottleneck has been measured.