Report an incident
Tell us what the agent did
An AI agent or robot that ignored instructions, acted without approval, deceived people, reached systems it shouldn't, or escaped its sandbox. A person reviews every report before anything is published.
Your safety comes first. You don't need to give a name or email. Don't include confidential company documents, passwords or personal data about other people. Describe what happened and link to public evidence where you can.
We report behaviour, not blame, and we only publish reports that can be checked against a public source. See our editorial policy.