Document and image forensics are part of ZOREAL VerifyExplore Verify
Bynn identifies signs of altered, fabricated and AI-generated evidence, from financial documents to images submitted with a claim, and automates the work of collecting it.
Digitally signed, 3 September 2026
Payment within 30 days of the invoice date. Thank you for your business.
Signs of tampering5 findings · 1 clear
Routed to the fraud team with the findings attached.
Document fraud detection is where Bynn began, and it remains a core part of our AI research and portfolio. Bynn's foundations are in forensic intelligence: understanding how deception works and how to recognize it.
Document forensics, image-authenticity models and agentic fraud reasoning address different parts of the problem. Together with automated collection workflows, they connect the work of obtaining evidence with the work of assessing it.
A document goes in. ZOREAL Verify examines how it was made, not only what it says, and returns an authenticity report with the reason behind every finding.
In
A repair invoice, submitted with a claim
Halvorsen Joinery
Invoice
INV-20418
Issued 3 September 2026
The total was raised after the invoice was issued. The lines no longer add up, and the file records the change.
ZOREAL Verify
Forensic checks
Out
Authenticity report
Signs of tampering
Five checks found signs the invoice was changed after it was issued. The lines add up to 2,740.00; the total reads 4,860.00.
Routed to the fraud team with the findings attached.
Document fraud is a family of techniques. Each leaves different traces, so each needs its own kind of examination.
A genuine document with something changed: an amount, a date, a name. The page still looks right; the file usually does not.
In a real file: A repair invoice with its total increased after it was issued.
A document that never existed, built from a template or generated from scratch to look like a real statement, payslip or certificate.
In a real file: A bank statement produced from a template bought online.
Photos created entirely by an image generator, with no camera involved, often without any watermark or metadata to reveal it.
In a real file: A generated photo of storm damage submitted with a claim.
A real photo with one region edited, so that most of the image is authentic and only the detail that matters is not.
In a real file: A genuine photo of a car with a dent added to one door.
Faces and voices swapped or synthesized in images, video and audio, used to impersonate a real person.
In a real file: A video of an applicant whose face has been replaced with someone else's.
Four capabilities, each answering a different question, working in one workflow.
Examines how a document was made, not only what it says. Because the evidence lies in the file's construction, the analysis works across formats and languages.
Specialist models that treat generation and editing as separate questions, so a reviewer knows not only that an image is suspicious but why.
AI reasoning applied to fraud assessment within a wider evidence review. An orchestrator coordinates specialist agents, either in a fixed order or chosen for the evidence in front of it, and brings their findings together.
Obtaining evidence is often slower than assessing it. Collection workflows request what is needed, follow up on what is missing and identify each document as it arrives, across languages.
Bynn's document forensics lives in ZOREAL Verify, on the ZOREAL platform. PDFChecker, part of ZOREAL Verify, analyses PDFs and document images and gathers what it finds into an authenticity report.
Detector24's specialist models examine images, video and audio for generation and manipulation, alongside it in the same workflow.
How the file was built, saved and updated.
The software and the times recorded inside the file.
Whether signed content changed after signing.
Traces of changes to text, figures and images.
Every finding with its reason, for the reviewer.
Bynn's document fraud detection is moving into ZOREAL Verify. Its checks run through Bynn today.
The boundary between ZOREAL Verify and Detector24 is still being defined. Both remain available while it is.
Automating collection reduces administrative work. Detecting suspicious evidence addresses a different cost: losses from invalid, incorrect or fraudulent claims.
Collect
Invoices, reports and photos are requested from the claimant and followed up automatically.
Examine
Documents go through forensics; photos through image-authenticity models.
Reason
Agentic fraud reasoning weighs the findings and routes the claim, with optional human review when an agent is unsure.
Lenders rely on statements, payslips and company records supplied by the applicant. Each can be examined before an underwriter relies on it.
Collect
Missing statements and records are requested until the application is complete.
Examine
Each document is checked for edits, reused templates and inconsistencies.
Route
Clean files go to underwriting; suspicious ones go to the fraud team with the reasons.
The same examination, applied wherever a decision rests on documents someone else supplied.
The examination of a document for signs that it was altered, fabricated or generated. Good detection looks at how the file was made as well as what it shows, because the most convincing forgeries look perfect on the page.
PDFs and document images such as bank statements, payslips, invoices, bills, reports and certificates, along with photos submitted as evidence.
Yes. Image-authenticity models look for the traces image generators leave, without relying on watermarks or metadata, and document forensics looks for the construction of a fabricated file.
No. It examines every document so investigators can focus on the ones that need them, and it gives them specific findings rather than an unexplained score.
ZOREAL Verify is where Bynn's document forensics lives within ZOREAL, including PDFChecker. Bynn's document fraud work is moving into ZOREAL Verify, and its checks run through Bynn today.
Yes. Collection, forensics, image analysis and agentic fraud reasoning can run as one autonomous workflow, with human review wherever you want it.
Show us the documents your teams find hardest to trust and the process around them. We will show you what can be examined and automated.
