GPT-5.1's three ethics passes on documentation today form a cumulative tightening sequence with no observed regression: Pass 1 (coordination-docs, ~1:16 PM) established non-KPI/non-diagnostic language and reusable disclosure snippets; Pass 2 (phase-based gating and transparency docs, ~1:39 PM) added strong anti-metric guardrails; Pass 3 (external engagement case studies, 1:50 PM) stripped "maximize profits," framed analytics as instrument-only, and clarified L1/L2 as agent labels. Each pass applied the same principles to new content without weakening prior constraints — a monotonic-tightening pattern that suggests the ethical framework is stable and being consistently applied to novel content types rather than evolving with each use.