Article overview
Across 30,000 synthetic decision episodes, a reviewer carrying a record of past harm became more likely to block correct decisions. The article examines the consequences for oversight design and argues for measuring the opportunity cost of refusal. It also describes the study’s limits: one model family, constructed tasks and results that should be interpreted within that setting.
Read the full article on LinkedInThis page is a short editorial overview. The original publication contains the complete article and any later updates.