According to Forbes, in September, Alameda County Superior Court Judge Victoria Kolakowski dismissed a housing dispute case, Mendones v. Cushman & Wakefield, after noticing a submitted video exhibit was an AI-generated deepfake. The video, labeled Exhibit 6C, featured a witness with a disjointed, monotone voice and a fuzzy, emotionless face that would twitch and repeat expressions. The plaintiffs sought reconsideration in November, arguing the judge only suspected but didn’t prove it was AI, but she denied the request. This case is among the first documented instances of a deepfake being submitted as authentic evidence in a U.S. courtroom, and judges across the country, like Minnesota’s Stoney Hiljus and Louisiana’s Scott Schlegel, say the system is completely unprepared. The U.S. Judicial Conference has proposed a new Federal Rule of Evidence 707 to address “machine-generated evidence,” but it’s open for comment until February 2026 and takes years to adopt.
The Trust Is Broken
Here’s the thing: this isn’t just about spotting a weird video. It’s about corroding the very foundation of evidence. Judge Erica Yew in California nailed it. What happens when someone uses AI to forge a vehicle title, gets a county clerk to file it, and then brings a *certified copy* to court? Suddenly, judges have to question sources that have been rock-solid for centuries. The whole “trust but verify” model is collapsing. As researcher Maura Grossman put it, we need to shift to “Don’t trust and verify.” But how do you verify when the detection tools are famously unreliable? It’s a nightmare scenario where the burden falls on judges who, let’s be honest, probably aren’t tech experts. They’re being asked to be forensic analysts overnight.
Slow Legal Meets Fast Tech
And that’s the core problem. The legal system moves at a glacial pace, while this technology evolves daily. A new federal evidence rule takes a minimum of three years to adopt, as retired judge Paul Grimm confirmed. By the time Rule 707 might be in effect, the AI tools will be ten generations ahead. So states are trying to go it alone, like Louisiana with its Act 250, which requires lawyers to use “reasonable diligence” to check if evidence is AI-generated. That’s a good step, putting the onus on the officers of the court. But it’s still reactive. Judge Schlegel’s point is crucial: the courts can’t do it alone. When a client hands you photos, you have to ask where they came from. Basic legal practice just got a lot more complicated.
A Harbinger Of Worse To Come
The personal examples from the judges are terrifying because they’re so plausible. Judge Schlegel said his wife could clone his voice with cheap software to fake a threatening message and get a restraining order against him. “You lose your cat, dog, guns, house, you lose everything,” he said. They’d grant the order every time. That’s the stakes. We’re not talking about a faked vacation photo in a divorce case. We’re talking about wrongful convictions, ruined lives, and the complete weaponization of “evidence.” The fear among judges is palpable, and for good reason. The proposed Rule 707 tries to apply the Daubert standard—used for expert testimony—to AI, requiring reliable methods and facts. But that framework is already clunky for humans; applying it to black-box algorithms is a whole new frontier of confusion.
What’s The Solution?
Look, there’s no easy fix. Awareness is the first step, and groups like the National Center for State Courts are trying to educate judges on the difference between “unacknowledged” deepfakes and “acknowledged” AI evidence, like a synthetic accident reconstruction. But detection tech is a shaky crutch. In the Mendones case, they got lucky—the video metadata said it was from an iPhone 6, which didn’t have the capabilities the story required. That’s a forensic tell that will vanish as generators get smarter. Basically, we’re in an arms race where the forgers have a massive head start. The legal system was built on the assumption that fabricating evidence was hard. That assumption is now dead. And until the tools and the rules catch up, every piece of digital evidence should be viewed with extreme skepticism. It’s a horrible way to run a justice system, but what’s the alternative?
