Rutgers Office for Research Helps Launch Company Using AI to Prevent Deepfakes
Steg.AI, a startup company based on artificial intelligence technology developed at Rutgers, is aiming to help businesses and organizations protect their media assets and intellectual property by making it easier to detect deepfakes – very realistic but phony visual or sound content – and determine if an item has been altered.
The technology, the brainchild of Wengrowski, who received both his undergraduate and doctoral degrees in electrical and computer engineering from Rutgers, was the result of a capstone senior project overseen by Dana who asked him to join her lab in the Department of Electrical and Computer Engineering.
“We leverage a research technology developed along with some of my cofounders called light field messaging, which is an advanced forensic water marking technique that adds information to files like images, video, pdfs, gifs, etc., that is invisible to us but visible to our algorithms or even a camera,’’ Wengrowski said. “This information is essentially embedded into these files as forensic tracers for our customers, so they can figure out who is doing what with their assets.”
Steg.AI’s mission is to establish a level of provenance for all digital media, using patented steganography technology to place attribution into content so that users can be sure that what they are engaging with is indeed real, or at least trustworthy.
The research conducted by Wengrowski and Dana, who tailored their research after meeting with various companies facing these issues, and their team focused on the ability to transmit information with light in a way that is only visible to a machine or a camera. This required an understanding of how light works and is captured by a camera and how the human visual system operates in terms of seeing and processing that information.
“We participated in the National Science Foundation (NSF)’s national I-Corps program and received feedback regarding the real-use cases of the technology, and we realized that we had a very compelling value proposition for information security,” said Wengrowski. “We learned that by talking to companies like Meta, Getty and Adobe, and then putting two and two together about how we could solve those problems.”
"We were able to leverage recent deep learning advances to build a robust solution to the pattern embedding problem,” said Dana. “By taking part in the Bay Area I-Corps in the Winter Cohort 2019, we were able to dedicate significant time and effort to in-person customer discovery to explore commercial needs for our technology."
This innovation could not come at a more critical time. According to The New York Times, the increase of content such as deepfakes and the ease in creating fake content using artificial intelligence has grown exponentially and will only expand further.
According to Wengrowski, current deepfake detection methods involve analyzing visual or aural content for artifacts that indicate it was created by an algorithm. Still, he feels that over time the algorithms will simply incorporate those detectors, and the fake content will pass through undetected.
Steg.AI, whose name stems from steganography, was able to parlay the technology into funding from a variety of sources. Besides the NSF I-Corps program, the company received non-dilutive funding from phase one and phase two NSF Small Business Innovation Research (SBIR) grants. According to Wengrowski, Steg.AI recently closed a seed fundraising deal with leading cyber security venture investors.
“Steg.AI is another example of Rutgers excellence and how the university is leading the way into the future,” said Deborah Perez Fernandez, acting executive director for Innovation Ventures. “The rise of deepfakes, mis- and disinformation highlights the importance of the potential impact that the technology developed by Dr. Wengrowski and Dr. Dana can have around the world.”
Rutgers Office for Research Innovation Ventures, the technology transfer department for the university, filed patent applications for the technology in the United States, European Union, China, Japan and India, and handled the execution of the exclusive license to Steg.AI.