Catch fake refund photos before they cost you.
AI-generated damage photos, synthetic missing-item evidence, and recycled claim images are costing retailers millions. Detect them before you issue the refund.
Refund fraud has gone synthetic.
Refund fraud costs retailers billions every year. AI image generators have made it trivial to produce convincing damage photos, and the same fake image can be submitted across dozens of accounts before anyone notices.
What fake claims look like.
"Damaged item" photos
AI-generated images of cracked screens, dented packaging, or broken products that never existed. Submitted with refund or replacement requests to get free merchandise.
"Missing delivery" evidence
Synthetic photos of empty porches, opened-but-empty boxes, or tampered packaging. Used to claim orders never arrived or arrived incomplete.
"Not as described" comparisons
Side-by-side fake photos showing a product that looks different from the listing. Generated to justify return requests on items that match exactly what was advertised.
Recycled photos across accounts
The same damage photo submitted by multiple accounts, sometimes with minor crops or filters applied. A single convincing image becomes a tool for serial fraud.
Screen every claim. Catch the fakes.
Manual review via web tool
Upload claim photos to the Witness web app for instant AI detection. Trust and safety teams get a confidence score and verdict in seconds. Available now.
API integration at claims submission
Integrate detection directly into your returns and claims workflow. Screen every submitted photo automatically before it reaches a human reviewer. Coming soon.
Duplicate detection across accounts
Perceptual hashing (pHash) identifies the same photo submitted across multiple accounts, even after cropping, resizing, or minor edits.
Route flagged claims to review
Clean submissions pass through. Flagged claims are escalated to manual review with full audit trail, confidence scores, and detection signals.
Built for the platforms that need it.
Common questions.
How does AI detect fake refund photos?
Witness analyzes submitted images with an ensemble of detection models. Each model checks for different signals: generative fingerprints left by AI image generators, frequency artifacts invisible to the human eye, and metadata inconsistencies. Results are combined into a single confidence score and verdict.
Can the same fake photo be detected across multiple accounts?
Yes. Perceptual hashing (pHash) compares submitted images against previously seen claim photos. Even if an image is cropped, resized, or slightly edited, duplicate detection can flag reuse across accounts and claims.
What types of e-commerce fraud can AI detect?
AI fraud detection screens for AI-generated damage photos, synthetic missing-delivery evidence, fake comparison photos used in not-as-described claims, and recycled images submitted across multiple accounts or platforms.
How do I integrate this into my returns workflow?
Today, trust and safety teams can use the Witness web app for manual review of suspicious claims. API integration for automated screening at the point of claims submission is coming soon. Contact us for early access.
Stop paying for fake claims.
Talk to us about screening refund and return photos with Witness.