PDF Tampering Detection

Checks each page of a PDF for edits and AI generation, and gives a verdict for the whole document.

Slug	ID	Status	Tagged	Model version
`pdf-tampering-detection`	`13`	Active	When a page was AI-checked	`v1`

What it detects

Each page is classified first, then checked:

Page	Check	Page verdict
A full-page scan with something added on top (visible text, annotations, filled shapes or an extra image)	The added items are marked	`EDITED`
A full-page scan with nothing added	AI image check	`FAKE` if AI-generated, `REAL` if not
Anything else (text, tables, logos or photos on a normal page)	Not checked	`OUT_OF_SCOPE`

It does not detect:

Pixel-level edits on a scan, such as Photoshop retouching with nothing added on top. These pages come out REAL.
Edits to native PDF text or tables. These pages are OUT_OF_SCOPE.
File-structure signs such as incremental saves or creator/producer mismatches.
Anything in encrypted PDFs, which are not analyzed.

Request

Slot	Accepted `contentType`
`original`	`application/pdf`

Send the PDF's page count as pageCount. It is required, and it sets the job's cost.

{
  "taskTypeId": "13",
  "inputs": [
    { "slotName": "original", "fileName": "agreement.pdf", "contentType": "application/pdf", "sizeBytes": 734003, "pageCount": 3 }
  ]
}

See Running a job for the full create, upload and finalize flow.

Limits

File size: up to the account upload limit, 10 MB by default.
Encrypted PDFs are not analyzed.

Result

Field	Type	Description
`verdict`	string	Document verdict; see below
`isFake`	boolean	`true` when `verdict` is `TAMPERED`
`totalPages`	integer	Pages analyzed
`fakePages`	integer	Pages the AI check found AI-generated (`FAKE`)
`editedPages`	integer	Pages with content added on top of a scan
`scannedPages`	integer	Pages that are full-page scans
`outOfScopePages`	integer	Pages that weren't checked
`aiCheckFailedPages`	integer	Scanned pages where the AI check couldn't run
`confidencePercent`, `fakeConfidencePercent`, `realConfidencePercent`, `tag`, `tagBoundaries`		Present when at least one page went through the AI check; see result fields
`modelVersion`	string	Version of the model that produced the result

`verdict`	When
`TAMPERED`	At least one page is `FAKE` or `EDITED`
`AUTHENTIC`	No tampered pages, and at least one page is `REAL`
`UNDETERMINED`	No tampered or authentic pages, and the AI check couldn't run on at least one page
`OUT_OF_SCOPE`	Every page is `OUT_OF_SCOPE`

{
  "isFake": true,
  "verdict": "TAMPERED",
  "confidencePercent": 82.9,
  "fakeConfidencePercent": 82.9,
  "realConfidencePercent": 17.1,
  "tag": "fake",
  "tagBoundaries": { "real": 30, "likelyReal": 50, "likelyFake": 75 },
  "totalPages": 1,
  "fakePages": 1,
  "editedPages": 0,
  "scannedPages": 1,
  "outOfScopePages": 0,
  "aiCheckFailedPages": 0,
  "modelVersion": "v1"
}

Use verdict to decide what to do with the document; an OUT_OF_SCOPE or UNDETERMINED document was not verified either way.

Output artifacts

`kind`	File	Description
`result`	`result.json`	Document verdict, page counts, SHA3-256 / MD5 hashes and file metadata
`result`	`result_N.json`	One per page: page verdict, the reason, any added items with their positions, and confidence
`heatmap`	`heatmap_N.png`	One per page where available: AI heatmap for scanned pages, added items marked in red for `EDITED` pages, or a plain render for `OUT_OF_SCOPE` pages

Download them from artifacts on GET /jobs/{id}.

Cost and availability

Priced per page: the per-page price set by your plan × pageCount. The exact amount is returned as cost when you create the job. See Quotas & Credits.
Available on the cloud API; your plan decides which task types you can use.