Document forensics
Two different forgeries. Two different detectors. One verdict.
What is document forgery detection?
Document forgery detection tells a genuine document from one that was altered or generated whole. Raid AI runs two detectors on a PDF: one boxes the region of a real document that changed, the other decides whether the page was generated from nothing, and the evidence from each is kept beside the single verdict.
Editing a genuine document and generating a fake one from scratch look nothing alike, and a detector tuned for one will confidently miss the other. Most document fraud tooling picks a side without saying so.
Raid AI runs both questions separately. One detector localises manipulation of a real document down to the altered region, which is what an investigator needs to see. Another decides whether the page was generated whole. Their outputs are fused into a single verdict that still carries the underlying evidence.
Onboarding and claims are where this lands: a queue of documents that a human cannot read closely at volume, where the cost of waving through a forgery is measured in regulatory exposure rather than inconvenience.
The attacks this answers
- Altered figures, dates and names on a real document
- Wholly AI generated pages presented as scans
- Reused templates dressed as fresh documents
- Identity documents modified after issue
From submission to verdict
Submit
Upload a PDF, optionally declaring the document type.
Localise
Manipulation of a genuine document is boxed to the region that changed.
Check for generation
A separate pass decides whether the page was generated whole.
Fuse the results
One verdict, with the evidence from each detector kept alongside it.
At a glance
- Input
- PDF upload, or fed by webhook
- Evidence
- Region level localisation of the alteration
- Method
- Layered detectors fused into one verdict
- Availability
- Enterprise
Questions about Document forensics
What kinds of forgery does it catch?
Two different ones, with two different detectors: an edit to a genuine document, where the altered region is boxed, and a document generated whole, which has no genuine original to compare against.
What does it take as input?
A PDF, optionally with the document type declared, so the right checks run for a bank statement, an identity document or a claim form.
How does it show its reasoning?
Every decision keeps an evidence trail: which detector fired, on which region, with what score. A reviewer can see why a document was flagged and defend the decision later.
Who can use it?
Document forensics is on enterprise plans, built for onboarding and claims teams.