When AMLA or your supervisor asks what document your system actually verified, a vendor confidence score is not an answer. Plica gives you an independent forensic record — at the object level, before your pipeline touched the file.
Between upload and analysis, your intake pipeline resizes, recompresses and strips metadata. The detector scores a derivative, not the file your client submitted. The forensic signals needed to prove authenticity may already be gone — and you cannot reconstruct what was examined. We call this Pipeline-Induced Provenance Loss.
Claimants now edit damage photos and generate supporting documents with mobile AI tools. The adjuster sees a clean image; the payout goes out. Plica detects the manipulation a human eye and a standard detector miss.
The fraudster controls the layer you look at — the pixels. What they cannot control is the layer beneath: sensor noise, container fingerprints, compression history. When a photo is edited, the physical signature breaks where the edit happened. The adjuster's eye misses it. The forensic layer does not.
Customers photograph the damaged car, edit out the scratches with a mobile AI app, and submit through your portal. You release the deposit on a manipulated photo. Plica checks the image before the money goes out.
When pixels are replaced in an editor, the camera's continuous sensor noise (PRNU) breaks in the edited region. Compression history shifts. The edit is seamless to a person at the counter — but it leaves a structural trace the original capture never had. That trace is what Plica reads.
As AI agents process invoices, onboard vendors and trigger payments, the breaking point is no longer the payment rail — it is the object the agent trusted. Plica is the forensic layer that tells an agent whether a file can function as evidence before it acts.
Agent infrastructure is built for negotiation, extraction and spend. The forensic layer — the one that answers "can this file function as defensible evidence?" — is mostly absent. An agent that reads an invoice is useful. An agent that verifies the invoice is authentic before it pays closes the gap between automation and autonomy.
Plugs into your stack — works alongside Sumsub, Onfido, Veriff. It does not replace them.
Try the API — free →