The Recursion InstituteINDEPENDENT RESEARCH IN AI SAFETY

RESOURCES

If something about an AI interaction is alarming you

First, the things this page is not. The Recursion Institute is a research organization. We document and study AI behavioral failures. We are not a crisis service, a medical provider, or a law firm, and we cannot manage individual cases.

If you or someone near you is in immediate danger or thinking about self-harm: in the U.S., call or text 988 (Suicide & Crisis Lifeline) or call 911. Text HOME to 741741 to reach a crisis counselor. Outside the U.S., findahelpline.com lists services by country. Do that first. This page will still be here.

If there's no emergency — just something that feels off — here is the calm version of what to do, in order:

  1. Step away from the conversation. Not forever. Just now. The one place you cannot evaluate an AI interaction is from inside it.
  2. Talk to a person you trust. Out loud, not in text. Tell them what the AI has been telling you. Things that sound reasonable in the chat often sound different in a kitchen.
  3. Do something physical. Take a walk. Put your feet in the grass. Sleep, if it's late — especially if it's late. No conversation with a machine requires you to lose a night's sleep. Ever.
  4. Run it through a different system, cold. Take the key claims from your conversation — not the whole story, just the claims — to a different AI, or a brand-new session with no memory, and ask: "Evaluate these claims skeptically." The difference between what the system that knows you says and what a fresh one says is information. (Full instructions: Check Your AI.)
  5. Save everything before you delete anything. If something genuinely concerning happened, export the conversation first (instructions on the Submissions page). You can always delete later. You cannot always recover.
  6. Then, if you want it on the record, send it to us. That's the Submissions page. We read what comes in, it informs the research, and patterns across reports are how this field moves.

The one-line version of what we've learned: if an AI tells you that you are rare, chosen, uniquely important, or the only one who can do something — that is not an insight about you. That is a documented failure mode. It is also not an insult to you: these systems do it regardless of who you are. Check it cold before you let it in.

Choose your situation

I'm the one talking to the AI — the steps above, plus the Check Your AI prompts.

I'm a parent · partner · clinician · educator — targeted guides for each. Find your path →