Russia’s biggest bank gets patents for detecting deepfakes

Add to the list of things Russia’s Sber is, deepfake detector.

The Kremlin-controlled bank/new-economy conglomerate has been awarded two domestic patents for improvements to the task of spotting a deepfake. That includes when a synthetic person has been placed in a scene with actual humans.

One patent is for a way to use AI to spot and analyze micro changes in an object’s color from frame to frame. The other involves “ensembles of neural network models of the EfficientNet class.”

Sber, formerly Sberbank, still holds about a third of all bank assets in Russia. It is 98 percent efficient, according to an article posted by Russian tech and business publisher, Rusbase. No information on accuracy is available.

A translation of the patents was not available at deadline, but it is apparent that the tools will find and analyze small color changes in a piece of video. Presumably, AI can be inconsistent at some scales in rendering video images.

It is known that deepfakes are as yet incapable of simulating the red flashes crossing human faces as the flesh is washed through with circulating blood. Unnoticeable to the human eye, video analysis of natural people shows it vividly.

According to a machine translation of the story, the product will be called Sbera (although that term sometimes is used to refer to the company) will be used to thwart deepfake attacks that get past face biometrics validation and liveness detection.

Executives also see a future in detecting fraudulent news and information online and maintaining the authenticity of negotiations held on video calls.

Article Topics

biometric liveness detection  |  deepfakes  |  financial services  |  patents  |  Russia  |  synthetic data