Preserving Evidence After Suspected AI Chatbot Harm

Chat histories can change or disappear; a defensible record requires more than screenshots.

What happened—and what the record establishes

When an AI conversation may be relevant to self-harm, fraud, harassment, or another injury, screenshots are helpful but incomplete. They may omit timestamps, system notices, deleted messages, account settings, model versions, edits, linked files, and surrounding context.

Preserve the device without altering it, export account data when safely available, retain emails and billing records, and document the app and version. Avoid public posting that exposes a minor’s records or private medical information. An attorney can send a preservation letter requesting server logs, safety interventions, moderation records, model routing, and relevant policies.

Preservation is not proof of liability. It protects information so experts and courts can later evaluate what occurred. Mental-health emergencies require immediate professional help; evidence collection should never delay safety.

Questions the law must answer

• Are messages stored locally, remotely, or both?
• Did the service permit exports, edits, or deletions?
• What safety warnings or crisis resources appeared?

Practical takeaways

• Keep native exports and original files, not only cropped images.
• Do not reset, trade in, or update a key device without advice.
• Call or text 988 in the U.S. for an immediate suicide or crisis concern.