Deepfake Videos Are Flooding Social Media – Can New Tool SAGA Help Stop Them?
A disturbing trend has emerged on social media platforms, where deepfake videos are being used for disinformation, social engineering, identity theft, and even recruitment scams. These extremely realistic AI-generated videos can be so convincing that even experts struggle to distinguish them from real footage. Now, researchers have developed a groundbreaking tool called Source Attribution of Generative AI (SAGA) videos, which aims to track down the source of these fake videos and help prevent their spread.
The SAGA tool is the brainchild of Rohit Kundu, a Ph.D. research intern at YouTube and doctoral candidate at University of California Riverside. Kundu’s team initially focused on developing AI detection models that could identify whether a video was fake or not. However, they soon realized that simply detecting AI-generated videos wasn’t enough – they needed to uncover the source behind them.
Using publicly available data, the researchers built a reasoning model that could explain why a video was fake and what specific tools were used to create it. The SAGA tool can identify the generative model used, its version, and even the development team responsible for creating it. This information can be invaluable in helping prevent the spread of deepfakes, as it allows researchers and platform owners to pinpoint the source of the fake videos and work with the model creators to improve their data filters and protections.
The SAGA tool uses a video transformer architecture that’s tailored specifically for video attribution. According to Kundu, this architecture makes it easy to adopt and deploy in real-world scenarios. Moreover, the researchers have proven its effectiveness by testing it on various AI-generated videos and achieving impressive results.
But how does the SAGA tool work? Essentially, it leaves a “temporal signature” (T-Sig) as it processes video data over time. This signature serves as a unique identifier that can be used to track down the source of the fake video. Kundu’s team has proposed that T-Sigs are the key to attribution in deepfake videos.
The SAGA tool is still in its early stages, and challenges lie ahead when it’s deployed in real-world scenarios. However, with its ability to identify the source behind AI-generated videos, it has the potential to revolutionize the fight against deepfakes. By sharing information about which models are being used to create fake videos, researchers can help model creators improve their data filters and protections.
As social media platforms struggle to keep up with the rising tide of deepfakes, tools like SAGA offer a glimmer of hope. With its potential to track down the source behind these fake videos, SAGA could be a game-changer in the fight against misinformation and identity theft. However, it’s crucial for users to remain vigilant and report any suspicious activity on social media platforms.
For now, the SAGA tool remains a promising development in the battle against deepfakes. But as Kundu himself notes, “If some big group’s model is being used to create phony videos, you’d want to let them know: ‘A lot of videos were generated using your technology, so you need to put more restrictions.'” The SAGA tool has the potential to be a powerful ally in this fight – and it’s up to us to ensure that we’re doing everything we can to prevent deepfakes from spreading.
Source: Dark Reading — 2026-08-03