Scam AI's technology
A frontier lab for scam prevention
We advance the state of the art in detecting AI-generated media. Eva V1.6, our unified detection API, is trained on global fraud data and keeps learning as new attacks emerge.
Built to hold up as attacks evolve
01
Multimodal by default
Face, document, voice, and device signals are scored together, not one channel at a time: an attack that beats one check, such as liveness, still has to beat the deepfake and document checks.
02
Retrained on a standing cadence
Eva is retrained as new attacks emerge. A detector that is not retrained loses about a third of its recall on new generators within about six months.
Static detector, day 0
~6 months
03
Evidence you can audit
Every result cites the exact signals that fired, backed by published, peer-reviewed research.
See the papers →