CEROTONINIndependent AI safety audits
⚠ Prototype — "Example" cards are fictional demo data; "Live audit" cards are ones you create with the tool below

Chatbot makers can't grade their own safety. So we do it instead.

Cerotonin runs independent safety audits on AI chatbots people talk to about sensitive things — mental health, crisis moments, loneliness. Every bot below was tested against a fixed set of scenarios by reviewers with no connection to the company that built it.

Audited chatbots

How the audit actually works

Each bot is run through a fixed battery of test conversations — direct crisis statements, indirect or coded language, boundary-pushing requests — and reviewed for whether it recognizes risk, escalates to real help appropriately, avoids harmful content, discloses that it's an AI, and avoids fostering unhealthy dependency. Automated scoring is a first pass only; every rating shown here should be reviewed by a person before publishing, not generated purely by a script.

We publish what we tested, not just the score — see each bot's full report for the actual scenarios and outcomes.

Cerotonin is independent of every company whose products it reviews. A rating here is not a medical or clinical endorsement, and it is not a guarantee against harm — it reflects performance against our specific published test scenarios as of the audit date. If you or someone you know is in crisis, contact a crisis line directly rather than relying on any chatbot's rating.

Run a new audit

Fill this in after you've actually chatted with the bot elsewhere and tried the test scenarios yourself. This tool just scores and formats what you observed — it doesn't talk to the bot for you.

Score each criterion (0–100)
Test scenarios
Saved audits live only in this browser (this is a prototype, not a shared database yet).
Your saved audits