This page was created on 2026-06-06 as a landing pad for a source assumed to be unidentified and un-ingested. The source is Anthropic, "Protecting the wellbeing of our users" (anthropic.com), saved at Raw Sources/Protecting the wellbeing of our users.md and classified source_class: supporting. Under the source-classification rules, a supporting source does not receive a Wiki/sources/ page; its content folds into the relevant model, concept, and company pages and cites as (Source: <URL>).
The post sets out Anthropic's safeguards for conversations involving suicide and self-harm — system-prompt guidance, reinforcement learning on human preference data, and a classifier that surfaces a support banner on Claude.ai — together with its work on reducing sycophancy and its 18+ age requirement. The evaluation figures it reports are cited on Claude Sonnet 4.6. Related coverage sits at AI Mental Health and Psychological Harm and Parasitic AI / Spiral Personas.
Retained as a record pending removal.
Relationships
- related: AI Mental Health and Psychological Harm — where this material belongs
- related: Claude Sonnet 4.6 — cites the post's evaluation figures directly
Sources
- Anthropic, "Protecting the wellbeing of our users": anthropic.com