Identifying whether content is real or fake on social media is becoming increasingly difficult due to the rise of AI. Watermarks are one solution; however, bad actors can remove them just as easily as they can make a political figure appear to say something lewd. That may be why researchers at Cornell University are proposing an alternative: coding the light in videos. This coding is invisible to the eye but embeds a hidden watermark that can be read to detect manipulation. From interviews and press conferences to images of buildings such as the United Nations Headquarters, this approach has broad applications. “Video used to be treated as a source of truth, but that’s no longer an assumption we can make,” said professor Abe Davis, who worked on the research project that led to the development of the technology. “Now, you can pretty much create a video of whatever you want. That can be fun, but also problematic, because it’s only getting harder to tell what’s real.” Despite the complexity of the technology, the developed Noise Coded Illumination (NCI) goes unnoticed by humans. The code consists of noise that naturally occurs in light, making it impossible to detect without knowing the secret code. Each coded light source has a unique code disguised in the noise of the video, creating information asymmetry through randomized illumination codes that resemble natural noise. In some cases, the coded videos showed a subtle flicker similar to fluorescent or LED lighting. But many were indistinguishable from the unedited versions. Image of man courtesy of iStock.com/simonkr. Magnifying glass image courtesy of iStock.com/roripond. NCI isn’t the first technology to combat fake media; others have made small changes to pixels to help viewers identify authentic footage. However, these approaches required specific cameras or AI models, which made integration more complicated for video creators. NCI ensures that any real video of a subject contains a secret watermark, regardless of who records it. To create this hidden watermark, the video is captured under special lighting, which can be programmed into computer screens, photography lamps, or built-in fixtures. Each watermarked light source contains a secret code that can verify the corresponding watermark in the video and reveal any attempts at malicious editing. Software for coding newer lights can be easily integrated, while older lights require a small computer chip attached to the fixture. Instead of using just one watermark, three separate codes can be embedded in the same scene, making it more difficult to fake. Forensic analysis can detect any gaps or cuts in the video, and if an object is added or replaced, the altered parts generally appear black in the recovered code videos. “Each watermark carries a low-fidelity time-stamped version of the unmanipulated video under slightly different lighting. We call these code videos,” Davis said. “When someone manipulates a video, the manipulated parts start to contradict what we see in these code videos, which lets us see where changes were made. And if someone tries to generate a fake video with AI, the resulting code videos just look like random variations.” NCI has been verified to work on footage captured outdoors and with people of various skin tones. This advanced video forensics approach could prove to be valuable in combating disinformation.