AI要約 Claudeに自作の計画書を監査させたところ、修正のたびに評価が反転し続けるシーソー現象が発生。AI監査の限界と、人間が判断軸を持つ重要性を実体験から考察した記事。

EN The author had Claude audit their own plan document, but each revision flipped the evaluation back and forth in an endless seesaw. The post reflects on the limits of AI auditing and why humans must hold the judgment axis.