Before founding Reality Defender, Colman held roles at Google, Goldman Sachs, and U.S. government-linked defense groups, all tied together by a common thread: “With all of these, there was an overlying theme, which was cybersecurity and data science, and really just kind of going to where bad people are using these for bad reasons.”
In an interview with The Software Report, Colman explains how Reality Defender detects deepfakes and AI-generated content across audio, video, images, and text. “All we’re doing is looking for statistical likelihood that whatever you’re seeing or hearing has indicated as AI generation or manipulation,” he says. The idea didn’t land at first. Over 200 VCs passed, and Colman recalls, “We came out of the woodwork saying we can detect AI avatars and virtual humans and the world said nothing because no one knew what that meant.” The core challenge, he explains, is that “the users of it do not and should not have to be that technical.”
The company first targeted high-risk environments like bank call centers. Detection now works across platforms—from Zoom to Teams to web browsers—and access is minimal. “Pick your favorite video conferencing or call center solution, either integrate with two lines of code or just go in the app store because we’re there, just click on it, two clicks of the mouse, and suddenly you have it.”
Behind the product is a serious technical engine. “There are times that we need to use 20,000 network machines for an evening to actually run an experiment,” Colman says, adding that they’ve “done a lot of creative things to reduce the amount of compute that’s needed.” The approach is backed by accuracy metrics. “For some of our largest clients that test us themselves, we have a less than 0.2% false positive rate,” he says. “Similarly, false negatives are probably even more important.”
As the team scales, hiring is a challenge. “There are no people who’ve been doing this for 10 years, 20 years. There’s people who’ve done this for 10 months in their PhD program,” Colman says. “I’d say only in the last year are we starting to see PhD dissertations actually on this.” Reality Defender’s culture is shaped by focus and purpose. “We’re not a charity because what we do is incredibly expensive in terms of GPU compute… but we also have a lot of fun, given that we’re working with AI,” Colman says. “Our team are really folks that are mission driven that don’t feel like they have to compromise their ethics or their morality.”
Colman sees a future where detection becomes part of everyday infrastructure. “You shouldn’t have to be a PhD researcher to identify something that’s fraudulent or fake,” he says. “It should just happen in real time.”