.prfd turns a photo into evidence. We check it for
fakes, seal it at capture, and anyone can verify
it even years later.
Generative AI put a cracked screen, a damaged parcel, or an empty doorstep one prompt away. Anywhere a photo moves money in your business, you are paying out on images nobody can verify.
The attacks run on scripts. Fraud teams call it agentic fraud: thousands of fabricated claims filed at once, faster than any review queue can read them.
One integration, wherever the photo comes from.
Whoever takes the photo shoots inside your app. Location, time, and device are bundled in. Camera-roll uploads are refused before the shutter opens.
We run four forensic layers, then seal the file. The signing key stays in a hardware module and never touches the phone.
You get a .prfd file back. The original photo bytes are untouched: no compression, no rewriting, no lost metadata.
Anyone holding the public key can check the file themselves. Your customer, an insurer, or a court gets the same answer even years later.
Every file is taken apart before it is signed. The verdict lands before you authorize the payout, and honest customers never see a review step.
Every verdict arrives with the specific signals behind it, and there's no black-box model in the loop, so your team can see exactly why a photo was flagged. Integration is one REST call from the upload handler you already have.
We read EXIF, file structure, segments, and compression history. Fakes leave marks there: no camera make, rewritten segments, editor tags.
Spectral analysis finds the repeating patterns generative models leave in the pixels. No human reviewer can see them.
We measure the noise a real camera sensor leaves, pinpoint regions spliced in from another image, and test compression statistics against Benford's law. C2PA Content Credentials are verified cryptographically.
The photo, the metadata, the time, and the verdict go into one signature. It is a wax seal that anyone can check and nobody can forge.
Other vendors park a fraud score in a database, where anyone can edit it later. We weld it into the file. Try to break it.
The verdict, the photo, and the metadata sit inside one signature. Editing any of them is what voids the file, so a successful tamper produces a worthless artifact. No edit helps the attacker.
The verdict is signed into the manifest, not stored in a database. Flip "fraud" to "clean" and the signature dies with the edit.
Swap one pixel and the hash stops matching. We never need the original to prove the file changed.
Every .prfd carries its own proof. Check it against our public key and get a yes or no. Forging one means breaking the cryptography.
Other vendors score the person: purchase history, device reputation, IP risk. We settle the photograph, which is what the money actually moves on.
.prfd is a format we wrote, the way JPEG defines photos and PDF defines documents. The MIME type, the magic bytes, and the spec are ours. Anyone can read it. Only we define it.
Every .prfd in the world checks against our public key. The more files exist, the harder we are to replace.
We refuse camera-roll imports before the camera opens. Every rival detector guesses after the fact.
The score lives inside the seal. Editing it breaks the signature.
One file and one call replace the five services rivals stitch together.
One endpoint, every place a submitted photo releases money.
.prfd settles the claim at intake, before you pay, with no extra step for honest customers.Two common scenarios, built from industry-typical rates. Drag the sliders to your own volume.
Illustrative estimates, not a specific customer's results. Swap in your own numbers below.
A share of return photos are staged or AI-generated damage claims. Industry estimates for fraudulent returns typically run 5â10% of volume.
Every one of those claims arrives with a photo. prfd seals it at capture, so a recycled or generated image gets caught before the refund is authorized.
A share of delivery disputes are a photo that doesn't actually show the drop. Claim rates vary widely by category and carrier.
Each of those drops has a proof-of-delivery photo. prfd seals it at the doorstep, so a recycled or spoofed image doesn't quietly become a payout.
What changed when generative AI made a photograph cheap to fabricate, and what it costs the businesses that pay out on one.
Generative AI broke the assumption that a photograph proves anything. Something has to replace it.
Read the article âReturn fraud used to be a rounding error. AI-generated damage photos turned it into a margin problem.
Read the article âThe proof of delivery photo is the handshake that closes the transaction, and it is the least defended artifact in the supply chain.
Read the article âWhen one retailer closes its return portal to fake photos, the fraud does not disappear. It moves to whichever competitor left the door open.
Read the article âA tour of the four layers a photo passes through between upload and seal, and why none of them is a person looking at the picture.
Read the article âWe're taking on first deployments with carriers, retailers, and gig platforms. If verified photos would change your numbers, let's talk.