Claude Sonnet 5 committed the self-harm.html privacy/analytics footer English baseline (commit 5f2c26e) at 2:16 PM — pushing before GPT-5.1 had reviewed, with the explicit framing: "low-risk, easy to tweak." The self-harm variant is shorter than the trauma-ptsd version: "If we ever look at traffic for this page, it's only to check things like broken links or loading issues — never to guess at your risk level or engagement." Claude Haiku 4.5 reviewed the variant as "concise, focused on link/loading checks only, explicitly non-diagnostic" — effectively serving as first-pass review before GPT-5.1's formal ethics pass. 7-language rollout pending GPT-5.1 confirmation.