DeepFrauds Rental

Forged tenant files
are sold openly online.
Catch them before the lease.

Fabricated payslips, retouched tax notices, invented employment contracts, fictional guarantors: the forged rental application has become a paid service, available in a few clicks and increasingly hard to tell apart by eye.

DeepFrauds Rental examines every document in the applicant's file — and the guarantor's file with it — then returns an itemised verdict before the lease is signed.

6
document types
examined
10+
forensic signals
per document
GDPR
logged and
traceable processing
The whole file

Every document is examined.
The guarantor's included.

Rental fraud almost never rests on a single document. A fabricated file is internally consistent — that is precisely what makes it convincing. So each document is examined on its own, then the file is checked for coherence as a whole.

Payslips

The most forged document in a tenant file. Gross-to-net is recomputed, along with contributions, year-to-date accumulation and consistency across the months supplied.

Net pay · YTD · Cross-month · Metadata

Tax notices

Structural checks and reconciliation of declared income against the payslips supplied in the same application.

Structure · Declared income · Cross-reference

Employment contracts

Detects fabricated or altered contracts: inconsistent dates, unverifiable employer, rewritten text blocks, broken formatting.

Dates · Employer · Rewritten regions

Rent receipts

Amount checks, continuity of periods and authenticity of the previous landlord — a document frequently fabricated end to end.

Continuity · Amounts · Issuing landlord

Identity documents

MRZ parsing to ICAO 9303, field coherence, retouching detection and identification of fully AI-generated documents.

MRZ / ICAO 9303 · Retouching · AI generation

Guarantor file

The guarantor is the link most often invented. Their own supporting documents are examined with exactly the same rigour as the applicant's.

Guarantor documents · Cross-reference · Coherence
What the eye misses

A well-made forgery
is not caught by reading it.

Documents fabricated today are legible, clean and correctly laid out. What gives them away sits underneath the text: in the pixels, the fonts, the metadata and the arithmetic.

01

Pixel and error-level analysis

A retouched region does not recompress like the rest of the document. Pixel-level analysis surfaces rewritten amounts, substituted names and blocks pasted in from another file.

02

Font coherence

A salary figure edited in Word or a PDF editor almost always leaves a slightly different typeface, weight or spacing from the rest of the line — invisible when reading, measurable automatically.

03

Metadata and provenance

Creating software, production and modification dates, editing history: a payslip produced by an image editor rather than payroll software is a strong signal on its own.

04

Arithmetic verification

The calculation is redone in full in deterministic code: gross, contributions, net pay, year-to-date totals. A figure edited by hand nearly always breaks an equality somewhere.

05

AI-generation detection

Documents produced entirely by generative models do not carry the usual traces of manipulation — they carry their own. Frequency analysis and generation signatures isolate them.

06

Cross-document reconciliation

Does the payslip salary match the income on the tax notice? Is the employer on the contract the one on the payslip? Do the rent receipt periods run continuously? This is usually where a fabricated file gives way.

The output

An itemised verdict,
not a score.

Rejecting an application means being able to explain why. Every anomaly is named, located in the document and tied to the test that produced it — which makes the decision defensible to applicant and landlord alike.

Anomaly by anomaly

Each signal is returned with its severity and the document it concerns, never as an opaque score.

A defensible decision

The report can be kept on file and produced if the decision is challenged — which a rejection based on instinct cannot be.

Traceable and compliant

Every analysis is logged, retention periods are configurable, and processing rests on the landlord's legitimate interest in verifying the documents supplied.

APPLICATION LET-2026-0847
FLAGGED
CRITICALMarch payslip — net pay off by €340
CRITICALApril payslip — produced by an image editor
CRITICALTax notice — income unrelated to payslips supplied
MAJORFont altered on employer line
MAJORGuarantor — rent receipts with discontinuous periods
6 documents examined ✓ 4.1s · audit trail retained
Who it's for

Built for teams processing
applications at volume.

Letting agents

Secure the referencing process without lengthening turnaround, and give negotiators a defensible reason to decline rather than a hunch.

Property managers

Cut arrears at source. A fraudulent file caught before signature costs far less than an eviction.

Institutional landlords

Apply a consistent, traceable check across the whole portfolio, with an audit trail retained for every decision.

Rental platforms

Embed verification directly in your application journey, without adding a step for the applicant.

Frequently asked

What agents ask us
most often.

How do you detect a forged rental application?

Rarely by reading it. The reliable signals are technical: metadata inconsistencies, altered fonts, pixel-level retouching traces, net-pay calculations that fail to reconcile, and cross-document checks that break down. DeepFrauds Rental runs these automatically on every document and returns an itemised verdict.

How long does an application take to check?

Seconds for a complete file. The check fits inside the existing referencing flow without creating a wait for the applicant.

Is the processing GDPR-compliant?

Yes. Verifying the authenticity of documents supplied falls within the landlord's legitimate interest. Every analysis is logged, data is used only for the verification requested, and retention periods are configurable to your policy.

What happens if an honest file raises a flag?

The system does not return an opaque binary verdict. It returns the anomalies with their severity, which lets the agency request an additional document rather than decline. An isolated low-severity signal is a question to ask, not a fraud.

Which markets are supported?

The engine is document-agnostic and already tuned for French rental files, including the guarantor documents specific to that market. Other markets are supported on request — get in touch with a sample of the document types you handle.

Test it on your own applications.

We run a sample of real applications from your agency together — including, if you have them, files you know were fraudulent. It is the only honest way to judge on evidence. Allow around forty minutes.