Courts Are Pulling Down Deepfakes. Is Your Video Evidence Next?

A fake video of Indian cricket coach Gautam Gambhir — showing him resigning from his position — racked up nearly three million views before anyone could stop it. Three million. And by the time a court stepped in, the damage was already done.

That story matters whether you've ever heard of

That story matters whether you've ever heard of Gambhir or not. Because the same technology that put fake words in his mouth can put them in yours. Anyone who's ever appeared in a photo online, posted a video, or sat on a work call with their camera on — this is about you now. On March twenty-sixth, the Delhi High Court ordered Meta, Google, and Amazon to pull down deepfake content linked to that hoax. The platforms had thirty-six hours to comply. That's a court — not a tech company, not a policy team — stepping in and saying synthetic media is a legal problem, not just a content moderation headache. And it raises a question that runs through everything we're about to cover. If courts are now treating fake video as something that demands proof of authenticity — what happens to all the real video that nobody's been proving is authentic either?

Start with what happened in Delhi. A deepfake clip showed Gambhir appearing to resign as head coach of India's cricket team. It wasn't real. But it spread across platforms so fast that nearly three million people watched it before any correction could catch up. Retroactive takedowns — pulling something down after it's gone viral — don't undo that kind of reach. The court's order was significant not because it solved the problem, but because it drew a line. Judges are no longer waiting for platforms to self-regulate. They're issuing deadlines.

And Delhi isn't alone in moving the legal goalposts. Across the U.S., forty-seven states had passed some form of deepfake legislation by mid-twenty-twenty-five. Forty-seven. On top of that, federal advisory committees proposed a new addition to the Rules of Evidence — Rule nine-oh-one-C — specifically designed to govern what they call "potentially fabricated or altered electronic evidence." In plain terms, that means before you can show a video or image in court, you may soon need to prove — with documentation — that it hasn't been tampered with. That's a massive shift. For investigators, it means building a paper trail for every visual they collect. For the rest of us, it means the next time a video surfaces in a lawsuit, a custody dispute, or a criminal trial, someone's going to have to answer the question: how do we know this is real?

Now widen the lens a bit. Facial comparison — the process of looking at two images and determining whether they show the same person — has been used in investigations for years. But according to the National Academy of Sciences, there's a serious gap. The Academy called for validation studies and error-rate measurement because, as they put it, there isn't enough evidence for the reliability of facial comparison methods as they're currently practiced. That means judges are starting to ask not just "what did you find?" but "how exactly did you find it, and can you prove your method works?" The recommended approach involves manual analysis of specific facial features — the shape of an ear, the spacing of the eyes — and then evaluating how common or uncommon those features are in the broader population. That's painstaking, documented work. It's the opposite of pulling up two photos side by side and saying "yeah, that's the same guy." And when automated systems get involved — algorithms that spit out a matching score — the credibility gap gets even wider. According to research published in forensic science journals, those automated systems lack methodological standardization and empirical validation in courtroom settings. Translation: the software gives you a number, but nobody's agreed on what that number actually means under oath.