Checking insurance claim photos for AI edits

AI editing tools can now paint damage onto an intact object or stage an accident scene. Claim photos need checking for when and with what they were taken, and for where they were changed.

Get the captured original

An original received through a claims app or email keeps grounds such as the capture device, time and compression history. A copy received through a messenger has lost most of this, and much less can be checked.

Mark the edited region

The alteration axis tells re-encoding, partial edits and screenshots apart and, for images, shows suspect regions as a heatmap. If only the damaged area shows a different compression history, that region is marked for the adjuster to look at.

Edits we still miss

Edits that erase or replace objects with the latest generative fill were rarely caught even in our own measurement. Do not read the absence of edit signs as a guarantee that nothing was changed.

Decide together with the record

Keeping the report's hash and ledger in the claim file makes it possible to reproduce which file was checked and what was found. The result supplements, and does not replace, site inspection and other documents.

Frequently asked

Is damage drawn with AI caught?
Edit traces are marked when present, but sophisticated generative edits are often missed.
Can several photos be checked at once?
Bundle them in a ZIP or send them through the API; each is scanned and the results are shown together.
Can the capture date be trusted?
EXIF dates can be rewritten, so they are read together with other grounds.
Are customers' photos kept?
Files are deleted after analysis; only the hash and report remain.

Read next

This is general guidance, not legal advice. A scan result is a probabilistic assessment, not proof.

Try a file