The deepfake battle is joined by Microsoft, banks and biometrics companies
Is it possible to always or at least reliably detect synthetic media? It seems unlikely today, much less as deepfake techniques and technologies continue to improve.
And yet researchers with a number of organizations say they are making progress.
Microsoft researchers say they have developed a product, the Video Authenticator, and unnamed software, part of Microsoft’s Azure, that allows certificates to be set in content and confirmed when the content is viewed.
Video Authenticator rates in real time the likelihood that photos and video frames have been adulterated. Microsoft says it detects the blending boundaries of deepfakes and “subtle fading or gray-scale elements that might not be detectable to the human eye.”
A trio of university researchers, meanwhile, say that they can both detect a deepfake and identify the specific generative model used to create it.
The researchers, from Binghamton University and Intel, figured out that residuals of generative models can reveal which model was used to build a deepfake. The residuals are analyzed against biological (and potentially one day biometric) traits of genuine humans – traits like heartbeat, breathing and blood flow. Deepfakes to date cannot recreate or approximate these exceedingly subtle signals.
The team claims 93.39 percent success on the FaceForensics++ data set detecting which of four generators and real videos were the sources of content samples analyzed.
And it is not only academics and technology companies taking a hard look at this problem.
UK-based financial services giant HSBC reportedly has decided to use a biometric identification product from Mitek Inc. and Adobe. It will be rolled out in the bank’s U.S. retail banking sites, analyzing live images and electronic signatures.
According to reporting by the Financial Times, the scheme is part of an ID verification product now being used by ABN Amro, Chase, Mastercard and other financial companies.
UK biometrics provider iProov found in a survey put out at the beginning of this year that more than three quarters of financial services companies’ cybersecurity leaders are concerned about deepfake-based fraud, but only slightly more than one quarter had put any measures in place to prevent it.
In response, iProov launched a new Security Operations Centre in Singapore earlier this month, which will spot and bounce attacks, including deepfakes. iProov provides biometric identity verification services for several bank customers including Rabobank and Knab.
This post was updated at 9:14am Eastern on Monday, September 28 2020 to clarify that the researchers are from Binghamton and Intel.