Fifty percent isa coin flip.
That is how often experienced reviewers spot AI-manipulated damage photos. Chainmail stamps photos on the driver's phone, so edits made after stamping show at review. A stamp cannot tell AI from camera.
Accidents that never happened.
Generative tools can add a dent, a scrape or a second car. Photos reused from other claims still slip through.
for experienced reviewers spotting AI-manipulated damage photos: no better than chance
Milliman, Dec 2025claims had a duplicate photo: 1,967 of 768,000 images
Verisk, Jun 2023in fraudulent motor, home and travel claims at Admiral in 2025
Admiral, 2025 figures (undated release)Guesses, stripped data, and trust.
Review by eye
Adjusters judge photos by eye at volume.
Duplicate checks miss edits
Reverse-image search finds reuse, not a changed dent.
Detectors guess
AI scores shift with every new model.
Stamp on the driver's phone.
Drivers send photos in Chainmail. The stamp is sealed before upload, and any later change fails the check. It shows edits after stamping; it does not prove a camera took the photo.
One chat
Photos and messages for a claim in one chat.
Edits show
Changed after capture? The check fails.
Checked locally
The check runs in the browser. Files do not upload.
Next steps in the chat
Send instructions where the photos are.
What ships today, and what is next.
Crash photo capture
With stamping on, every JPEG from the app is stamped on the phone.
Intact or changed
The check shows intact, changed or no stamp, with its limits.
Verified adjusters
A verified profile for your claims team.
Example brand. Verified profiles are not live yet.
Dash-cam and video
Stamped clips for the moment of impact.
Four steps.
First notice
The driver opens the claim chat from a link.
Capture
Photos are stamped on the phone before upload.
Check
Intact, changed, or no stamp.
Settle
Review files that fail or have no stamp first.
Where a stamp helps.
Every result states its limits.
- The JPEG has not changed since it was stamped.
- The fingerprint matches a record on Solana, with its time. Devnet today; mainnet is in development.
- Anyone can repeat the check with open-source code.
- That the photo was taken in Chainmail. Imported photos can be stamped too.
- Who took the photo, or who sent it.
- That a camera made it rather than an AI model.
- That the scene was honest when the photo was taken.
See it on your flow.
Coming soon to the App Store. For teams, we set up a demo on your own photos.