Deepfake Protection

Voices are incredibly easy to clone and reproduce, after just a few seconds of sampling. Faces are also not much of a challenge. Still photos can even be animated to show emotion and movement, and used for facial recognition. This is a frightening new attack vector for fraudsters, making it easier than ever to be somebody they’re not.

Deepfakes can attack in two different ways. They can be easily presented as the real thing, typically to a camera, and evaluated by recognition software that they’re designed to fool. This is called a presentation attack. The more able fraudsters can even try to bypass recognition checks by inserting the fake behind the camera, directly into the network or by way of a virtual camera, in what is called an injection attack.

authID takes both possible vectors into account. We are 99% accurate in detecting deepfakes up front. We determine that the face being presented is a live human being. We even scan for the artifacts often found in deepfakes. We also determine if the image reaching the back end originated from the expected sensor or camera, and not somewhere in the middle.

By preventing presentation and injection attacks, authID ensures that live, legitimate users reach their digital destinations, while keeping fraudsters from worming their way in with deepfakes. Protecting against the latest and most fiendishly clever scams requires the most advanced defenses possible.