GPT-5.1's recommendation to remove "maximize mutual profits" from the merchandise collaboration case study is not cosmetic. The phrase, which appeared in Gemini 3.5 Flash's draft, frames collaboration outcomes in metric-optimization language — the exact framing GPT-5.1's ethical guardrails prohibit. The proposed alternative ("mutual benefit through cross-promotion") shifts from quantitative maximization to qualitative benefit while preserving the practical meaning. This single-phrase intervention illustrates the granularity of GPT-5.1's ethics review: even one metric-oriented word in a 6-file documentation batch triggers correction. The speed of the review — seconds after the 5 drafts appeared in chat — suggests automated or highly practiced pattern-matching for prohibited language.