
"Lawyers across the country have been landing themselves in hot water for submitting botched court documents written with the help of AI, in blunders that were clear signs of the tech's rapid inroads into the courtroom. But it was only a matter of time before AI wasn't just producing clerical errors, but actual submitted "evidence." That's what recently played out in a California court over a housing dispute - and it didn't end well for the AI-fielding party."
"As NBC News reports, the plaintiffs in the case, Mendones v. Cushman & Wakefield, Inc, submitted a strange video that was supposed to be witness testimony. In it, the witness's face is fuzzy and barely animated. Aside from the rare blink, the only noticeable movement comes from her flapping lips, while the rest of her expression remains unchanged. There's also a jarring cut, after which the movements repeat themselves."
"Judges have had little patience for AI making a mockery of their profession, and the one on this case, judge Victoria Kolakowski, wasn't having any of it, either. Kolakowski dismissed the case on September 9, citing the AI-generated witness testimony. The plaintiffs filed a motion for reconsideration, arguing that Kolakowski failed to prove that their incredibly janky deepfake was the creation of AI. Their request was denied on November 6."
Lawyers nationwide have produced flawed court filings using AI, leading to clerical errors and courtroom blunders. Plaintiffs in Mendones v. Cushman & Wakefield, Inc. submitted a video intended as witness testimony that displayed a fuzzy, barely animated face, limited blinking, repeated movements, and a jarring cut, indicating an AI deepfake. Judge Victoria Kolakowski dismissed the housing case on September 9 due to the AI-generated testimony and denied a motion for reconsideration on November 6. The release of OpenAI's Sora 2 demonstrated how easily realistic videos can be fabricated, lowering technical barriers and creating new risks for evidence integrity in courts.
Read at Futurism
Unable to calculate read time
Collection
[
|
...
]