Full English policy
# PeakPact Elite — Responsible AI Policy **Effective date:** 15 September 2026 ## Principles 1. **Supportive, not coercive** — no shame, guilt, threats, or punitive language in automated coaching 2. **Minimized context** — send only fields needed for short motivational copy 3. **Human agency** — users can disable automation and ignore suggestions 4. **Safety fallbacks** — reject unsafe model outputs; use localized templates 5. **Transparency** — disclose AI involvement and source when shown in product 6. **No high-stakes sole automation** — AI does not independently ban users, charge cards, or adjudicate legal rights 7. **Locale respect** — generate in the user’s language; honor RTL presentation in clients 8. **Continuous review** — update prompts and denylist patterns as misuse patterns emerge ## Prohibited model uses - Medical, psychiatric, or crisis intervention advice - Generating deepfakes or non-consensual imagery - Discriminatory targeting - Manipulation of minors ## Governance Product owners review automation compose prompts, fallback catalogs, and incident reports quarterly (or after material incidents).