Safety
How moderation works across prompts, outputs, reports, and publication.
Before Generation
Prompts, character profiles, world and scenario definitions, story instructions, tags, titles, chat messages, and media references are checked before model calls when they can influence generated content.
Blocked requests receive a short refusal and a safe redirection instead of being sent to the model.
After Generation
AI chat output is checked before it is stored as the assistant reply. Unsafe output is replaced with a safe response.
Generated media should remain private or quarantined until output checks pass. Video review may include frame sampling, transcript checks, and OCR checks.
Operations
Safety events and reports create audit records with category, severity, surface, model/provider, decision, and timestamps.
Severe reports are routed to restricted moderation queues and legal escalation procedures where required.