Built to be the neutral number in an arms race.
Synthetic media gets better every month, and the detectors meant to stop it fall behind while their published numbers go stale. Margen exists to give buyers and vendors an independent measurement they can act on, with no detector of our own to sell.
An independent red team for deepfake detection.
We measure how well detection systems catch AI-generated media under the adversarial conditions buyers actually face, not the controlled conditions vendors test in their own labs. We have no detector of our own to sell. We are the measurement layer between the people who build detection and the people who depend on it.
Private evaluations
We red-team the detection a vendor or buyer brings us and hand back where it holds, where it fails, and the recipe behind every failure.
An open benchmark
We publish independent results on the detectors in the market, so buyers and vendors can check our work instead of taking our word.
Margen is the measurement layer between detection vendors and the buyers who depend on them.
An independent assessment layer.
Third-party, with no detection product of our own to sell.
A reproducible methodology.
Every claim is backed by a dataset, a pipeline, and a margin of error.
Adversarially honest.
We test under the hardest conditions buyers face in production.
Three commitments the measurement layer cannot exist without.
01 / Coverage
Coverage that tracks the threat.
Our evaluation corpus expands toward the frontier generators adversaries are adopting, so the benchmark keeps pace with the attack.
02 / Method
Reproducible methodology.
Every claim is backed by a dataset, a documented pipeline that recompresses media the way platforms do, and a pre-registered statistical methodology. Results can be independently re-run by anyone with corpus access on request.
03 / Fit
Context-fit evaluation.
We tailor the assessment to the threat the enterprise actually faces. The methodology is rigorous within each context, not generic across them.
An evaluation report has more than one reader.
Bring us a detector to measure.
Whether you build detection or depend on it, the value is the same: an independent number you can defend.