NewThe detectors that scored perfect collapsed the hardest under attack.

The best fakes already slip past your detector.

At scale, your detector is the fraud-prevention layer. We attack it with the fraud already circulating and show you where it breaks, before an attacker does.

Request an evaluationExplore the attack-data API
CLEAR
Reality Defender
NeuralDefend
RTscan
Slop Or Not
St Joseph's College of Engineering
CLEAR
Reality Defender
NeuralDefend
RTscan
Slop Or Not
St Joseph's College of Engineering

Why strong detectors still let fraud through.

A real face and its AI impersonation of the same identity. Drag to compare.

Trained on yesterday.

New generators ship monthly. Yours has never seen them.

Tested in a lab.

Detectors are never tuned for the platform conditions your traffic actually arrives in.

Measured on the average.

A strong average hides the subgroup that is wide open.

The detectors that scored perfect collapsed the hardest.

Clean lab test

100%

Decay

Real conditions

34%

Source: Margen open-source detector benchmark · 14 detectors

Read the benchmark

Comprehensively evaluate your detector on the conditions that fit your use case.

See all data cards
Passport and ID Face-Injection sampleHIDREAM
MG-KYC-01Available
Passport and ID Face-Injection
Remote KYCPassport / IDHiDreamQwen
2,316
real references
4,632
attacks
2
generator models
Open card
Synthetic Social Media sampleSDXL
MG-SYN-01Available
Synthetic Social Media
Social mediaSynthetic facesSDXL
27,997
clean images
18
conditions
12
demographic cells
Open card
Coming soon
MG-SWP-01Coming soon
Face-Swap Video
Face swapVideo
Notify Me
Coming soon
MG-PUP-01Coming soon
i2v Puppeteering
PuppeteeringImage to video
Notify Me
Coming soon
MG-LIV-01Coming soon
Active Liveness PAD
Active livenessPAD
Notify Me

Two ways to work with us.

Evaluation

We red-team your detector and help improve it.

Request an evaluation

Self serve

Pull off-the-shelf attack data through our API.

Explore the attack-data API

For the teams whose detector has to hold.

Continuous measurement of the detection layer in liveness and document checks.

What's included

  • Live-pipeline coverage. Adversarial samples evaluated against your real ingestion path.
  • Emerging-attack coverage. Tested against the attacks adversaries are adopting now.
  • False-positive control. Reduce false rejections that hurt real customers, without weakening fraud coverage.
Request an evaluationRead the full brief

Find your blind spot before someone else does.

Tell us what matters to you, and we will break down what needs to be measured.

Request an evaluationSee how we engage