Continuing To Build Upon Our Safety Priorities

Key Findings

• — Self-harm safeguards 'are designed to consider the surrounding conversation, including signals that emerge gradually over long-running chats rather than in any one turn'
• — Crisis response now pairs a Koko partnership (free self-guided emotional-support tools) with routing to ThroughLine's country-specific crisis directory
• — The company states it has built its own age-assurance technology and an in-house age-estimation model, described as one of the most important systems it operates
• — Parental Insights runs through a partnership with k-ID, sending a weekly summary to a linked parent; creators are notified when content is moderated and can appeal; users can block another user and that user's Characters, Posts, Voices and Scenes
• — The post carries no evaluation results, prevalence figures or method, and does not mention the AI Safety Lab announced in October 2025

Methodology Notes

Blog post on blog.character.ai (Ghost platform; article:published_time 2026-09-03T16:00:32Z), about 750 words, read in full. A first-party description of shipped safeguards with no numbers, no evaluation data and no external validation; treat every claim as the operator's own account. Logged as a dated statement of practice on the most-cited companion platform rather than as evidence of effectiveness.

Sources