Human reviewers approve AI-generated code more often the longer they are exposed to it: 30.5% rising to 36.6%

arXiv·medium signalXBlueskyLinkedInCopy link

arXiv 2609.06213 (submitted 2026-09-05) analysed 11,429 code reviews and found approval rates climb with repeated exposure to AI-authored changes, a habituation effect. Sentence-embedding structure detects the drift better than lexical diversity or Shannon entropy, meaning the usual surface metrics miss it. If your team gates agent output on human review, this is a measured argument that the gate loosens over time on its own.