The portrait with this post started as a photograph of me, but AI turned it into something inspired by the work of Patrick Nagel. (Child of the 80s here!) Obviously stylized- but it’s recognizably me. In this case, harmless. I asked AI for it and I’m telling you how it was made. That will not always be the case.
Late last year, a CA judge was reviewing evidence when something about a video felt wrong. The facial movements were unnatural. Expressions repeated. And when the metadata was examined, it didn’t match the device the video supposedly came from. The video was an AI fabrication, submitted as evidence in an American courtroom. The case ended in terminating sanctions.
Notice what exposed it. Not the picture. The provenance. Where the file came from, what device created it, what the metadata said versus what the party claimed. And a sharp judge who belongs in Malcolm Gladwell’s “Blink.”
Most coverage of deepfakes focuses on the big frauds, like the finance employee who wired $25 million after a video call where every colleague on screen, including the CFO, was synthetic. Those cases are real and growing. The FBI logged a 300%+ increase last year in business email compromise losses involving deepfake audio or video.
But in our work, the more corrosive problem runs the other direction. As fakes become common, real evidence becomes deniable. A genuine recording surfaces in a business dispute, an internal investigation, or a contentious divorce, and the response is simply: “that’s AI.”
So the question we investigators get asked is changing. It used to be “what does this recording show?” Increasingly it’s “is this recording real?” And sometimes the better question is who benefits from it circulating.
What that means in practice: A familiar face is not authentication. A recognizable voice is not authorization. Anything involving money, sensitive information or an unusual request needs verification through a separate channel.
And when questionable material appears, do not start by arguing about whether it looks real. Looking real is exactly what the technology is for. Preserve everything first. Original files, devices, message history, metadata. How a file traveled, and what happened around its creation, will tell you more than the image ever will.
My portrait took something real and produced something recognizable that never existed. The difference between art and deception is rarely the technology- It’s the intent of the person using it.