It's easier than ever to manipulate video footage to deceive the viewer and increasingly difficult for fact checkers to detect such manipulations. Cornell University scientists developed a new weapon in this ongoing arms race: software that codes a "watermark" into light fluctuations, which in turn can reveal when the footage has been tampered with. The researchers presented the breakthrough over the weekend at SIGGRAPH 2025 in Vancouver, British Columbia, and published a scientific paper in June in the journal ACM Transactions on Graphics.
“Video used to be treated as a source of truth, but that’s no longer an assumption we can make,” said co-author Abe Davis, of Cornell University, who first conceived of the idea. “Now you can pretty much create video of whatever you want. That can be fun, but also problematic, because it’s only getting harder to tell what’s real.”
Per the authors, those seeking to deceive with video fakes have a fundamental advantage: equal access to authentic video footage, as well as the ready availability of advanced low-cost editing tools that can learn quickly from massive amounts of data, rendering the fakes nearly indistinguishable from authentic video. Thus far, progress on that front has outpaced the development of new forensic techniques designed to combat the problem. One key feature is information asymmetry: an effective forensic technique must have information not available to the fakers that cannot be learned from publicly available training data.
