Infographic · Fraud & Document Assessment
Authentic vs. forged: what a reviewer actually checks
Forged documents rarely fail on one dramatic tell. They fail on accumulation — three small inconsistencies that each look survivable alone. This is the checkpoint sequence, and the logic behind the order.
This is awareness and detection material — the same category of guidance issued by issuing authorities, banks, and regulators to train front-line reviewers. It covers what a checker looks for, not how documents are produced. Nothing here is specific to any one employer's internal tooling or thresholds.
01The four layers of a genuine document
Every legitimate ID is built in layers, and each layer is expensive to reproduce for a different reason. A forgery usually clears one or two and quietly fails the rest.
02Side by side
- Optical features shift correctly as the document tilts
- Microtext stays crisp and legible under magnification
- Typeface, weight, and kerning match the issuer's standard exactly
- Photo sits within the substrate, under the same laminate as the print
- Encoded data agrees with printed data on every field
- Background line work is continuous, sharp, and uninterrupted
- Edges and corners are uniformly finished and sealed
- Personal details agree with the record on file
- "Optical" feature is flat — printed or foiled, doesn't respond to angle
- Microtext dissolves into a grey smudge when magnified
- Font substituted, or spacing subtly off from the issuer standard
- Photo edges lift, ghost, or sit above the laminate plane
- Encoded zone contradicts the printed fields, or is padded/malformed
- Line work breaks, blurs, or repeats where it should be continuous
- Edges delaminate, or corners show re-sealing
- Details plausible on their own but inconsistent with the record
03The checkpoint sequence
Order matters. The sequence runs cheapest-and-fastest first, so that obvious failures are caught before anyone spends time on deep inspection — and so that a genuine document reaches a decision quickly.
Frame the document
Confirm the document type, issuing authority, and version are what they claim to be — and that this issuer actually issues this format.
Check optical behaviour
Features that depend on angle or light source. A flat reproduction of a dynamic feature is one of the highest-signal tells available.
Magnify the fine print
Microtext and background line work. These are resolution-bound: they survive genuine production and degrade under copying.
Interrogate the typography
Issuers are rigid about type. Substituted fonts and drifting spacing are the most common failure in otherwise careful work.
Inspect the photo integration
Whether the portrait is part of the document or sitting on top of it. Tampering concentrates here because it's the field worth changing.
Cross-check encoded vs. printed data
Machine-readable zones, barcodes, and chips should tell the same story as the front of the card. Forgeries often update one and forget the other.
Reconcile against the record
The document can be entirely genuine and still be the wrong answer — if it doesn't match the person or the file it's attached to.
A single anomaly is a question, not a verdict. Escalate on accumulation and category: three soft flags across independent layers is a stronger signal than one hard flag that might be wear and tear. Document the reason code every time — an unexplained rejection is as much a process failure as a missed forgery.
04Where reviewers actually go wrong
| Failure mode | What it looks like | Counter-measure |
|---|---|---|
| Anchoring on one feature | Reviewer confirms the hologram, then stops looking | Run the full sequence regardless of early confidence |
| Fatigue drift | Accuracy quietly degrades late in a shift or queue | Rotate reviewers; sample-audit late-queue decisions |
| Wear-and-tear excuse | Real damage used to explain away genuine anomalies | Damage explains degradation, never contradiction |
| Volume pressure | Throughput targets silently outrank accuracy | Measure both; never publish one without the other |
| Undocumented rejects | Correct call, no reason code, unusable for appeal or audit | Reason code mandatory before the decision commits |
05The pressure the role is under
Detection is not a static problem. Published industry data puts AI-assisted document forgery at roughly 2% of detected fakes in 2025 — up from effectively zero the year before — while sophisticated attacks rose sharply as basic ones stopped working. The reviewer's job gets harder precisely because the controls work.
Sources
- Sumsub — Identity Fraud Report 2025–2026
- Regula — Identity Fraud by Numbers: Trends, Insights & Threats
- World Economic Forum — How identity fraud is changing in the age of AI