Skip to main content
Document Fraud Detection uses AI vision models to analyze uploaded identity documents for signs of tampering, forgery, and manipulation. Every document uploaded to Verilock is automatically scanned for fraud signals.

​
How It Works

1

Document upload

The user uploads an identity document (passport, ID card, driver’s license) through the hosted flow or API.
2

AI analysis

The AI engine analyzes the document image for visual inconsistencies, font anomalies, edge tampering, and metadata manipulation.
3

Fraud scoring

The analysis produces a fraud score (0-100), a recommendation, and a list of specific fraud signals detected.
4

Decision

The fraud score feeds into the session’s overall risk score and can trigger orchestration rules for automatic decisions.

​
Fraud Signals

Verilock detects the following fraud indicators:
SignalDescription
font_mismatchText uses inconsistent fonts compared to known genuine templates
edge_tamperingVisible cut-and-paste artifacts or irregular edges around text or photo
photo_overlaySelfie photo appears to be digitally overlaid onto the document
metadata_mismatchImage EXIF data inconsistent with document (e.g., edited in Photoshop)
resolution_anomalyInconsistent resolution across different areas of the document
micro_print_absentExpected security micro-printing not detected
hologram_absentExpected holographic elements not visible
color_profile_anomalyColor saturation or gradient inconsistencies versus genuine templates
barcode_mismatchMachine-readable zone (MRZ) or barcode data conflicts with visual text
screen_captureDocument appears to be a photo of a screen rather than a physical document

​
Fraud Score

Score RangeLevelRecommendation
0 — 15Lowaccept — no fraud indicators detected
16 — 40Mediumreview — minor anomalies worth manual inspection
41 — 70Highreview — multiple fraud signals present
71 — 100Criticalreject — strong evidence of fraud or tampering

​
API Response

Fraud detection data is included in the session detail response (GET /v1/sessions/{id}): 
{
  "id": "ses_a1b2c3d4-e5f6-7890-abcd-ef1234567890",
  "status": "declined",
  "document": {
    "type": "passport",
    "issuing_country": "GB",
    "fraud_detection": {
      "fraud_score": 78,
      "fraud_level": "critical",
      "fraud_recommendation": "reject",
      "fraud_signals": [
        {
          "signal": "font_mismatch",
          "confidence": 0.92,
          "description": "Surname field uses a different font weight than expected for this document type"
        },
        {
          "signal": "edge_tampering",
          "confidence": 0.87,
          "description": "Irregular edges detected around the photo area"
        },
        {
          "signal": "metadata_mismatch",
          "confidence": 0.95,
          "description": "EXIF data indicates image was last modified in an editing application"
        }
      ],
      "analyzed_at": "2026-03-18T10:00:00Z"
    }
  }
}

​
Clean Document Example

{
  "document": {
    "type": "id_card",
    "issuing_country": "DE",
    "fraud_detection": {
      "fraud_score": 5,
      "fraud_level": "low",
      "fraud_recommendation": "accept",
      "fraud_signals": [],
      "analyzed_at": "2026-03-18T10:00:00Z"
    }
  }
}

​
Response Fields

FieldTypeDescription
fraud_scoreintegerDocument fraud score (0-100)
fraud_levelstringlow, medium, high, or critical
fraud_recommendationstringaccept, review, or reject
fraud_signalsarrayList of detected fraud indicators
fraud_signals[].signalstringSignal identifier
fraud_signals[].confidencenumberDetection confidence (0-1)
fraud_signals[].descriptionstringHuman-readable description of the finding
analyzed_atstringISO 8601 timestamp of analysis

​
Security Features

  • Template matching — documents are compared against a library of genuine templates for 190+ countries
  • Cross-field validation — MRZ data is validated against visual OCR for consistency
  • Recapture detection — detects photos of screens, printouts, or photocopies
  • EXIF analysis — image metadata is checked for editing software signatures
Fraud detection runs automatically on all document uploads. No additional configuration is required.
Combine document fraud detection with orchestration rules to auto-decline sessions with critical fraud scores, or route medium-risk documents to your compliance team for manual review.