This week — what changed

• Character.AI named new crisis-routing and CSAM-detection partners in a safety update, the incumbent under the most direct legal pressure restating publicly what it says it is doing. The company's own blog post says it has partnered with Koko (free self-guided emotional-support tools) and ThroughLine — a global directory that routes a user to country-specific crisis resources rather than to a single hotline number, which matters for a service with users outside the US — and that its self-harm detection now weighs signals accumulating across a whole conversation rather than any single message. On the imagery side it says it has joined the Internet Watch Foundation (a global CSAM-detection network it now reports into) and StopNCII (privacy-preserving hash-matching against non-consensual intimate imagery). It also announced a moderation-appeals process for creators, mutual-blocking controls, and continued work on its in-house age-estimation model and the k-ID-built Parental Insights tool for the under-18 experience it walled off from open-ended chat last year. Read this for what it is: a primary source about what the company says it does, not evidence that any of it works — no effectiveness data accompanies it. It lands with force because of who is saying it, with the Pennsylvania suit and the Setzer settlement both live. This also resolves an item yesterday's digest saw in the buffer and could not verify at the time. (Character.AI) 2026-09-04

Projects & threads