Video: Woman’s mouth did not match the words being spoken
As fake lawsuits generated by artificial intelligence (AI) plague courts around the world, judges in the United States are facing the next frontier: deepfake videos submitted as evidence.
California Superior Court Judge Victoria Kolakowski issued a “termination sanction” effectively canceling the lawsuit after determining that two videos submitted by the self-proclaimed plaintiffs seeking summary judgment were fake.
She wrote that the characters in the videos (watch one of them here) are “inexpressive, monotonous, don’t pause where a pause is needed, make strange word choices, and generally seem robotic.” “Additionally, the mouth flaps do not match the words being spoken.”
The judge also cited a “loop in the video feed” as another reason to believe the video was the product of generative AI.
At an earlier hearing, she said, “The court’s suspicions were further strengthened by the statements made by Maridol Mendonez.” [one of the claimants] She said some of the witnesses described in the suspect’s evidence submission have died or been unable to contact her. ”
She also identified other evidence that wasn’t as good but had been “substantially altered,” including a photo allegedly taken with a Ring doorbell camera. “Upon closer inspection, it was determined that the background was black and white, but the man was in color,” the judge wrote.
She suspected more evidence had been tampered with, but said she did not have the “time, money or technical expertise” to determine whether that was the case.
Judge Kolakowski decided not to pursue criminal charges against the plaintiffs, saying that the remedy was both “too harsh and not sufficient,” and determined that termination sanctions were appropriate.
“This sanction is proportionate to the harm caused by Plaintiff’s abuse of the court process. After a significant breach of trust between the Court and Defendants, the termination sanction has the appropriate remedial effect of denying Plaintiff, and other litigants who seek to utilize GenAI to submit video testimony, the ability to further prosecute this case.
“Furthermore, the termination sanction has the appropriate deterrent effect of demonstrating to the public that the court will have zero tolerance for attempts to introduce deepfakes as evidence. The sanction serves as an appropriately chilling message to litigants appearing before this court: Please exercise extreme caution when using GenAI in court.”
Judge Scott Schlegel, a member of the Louisiana Fifth Circuit Court of Appeals, said in a blog about the case that he had been warning this day was coming for some time.
He acknowledged that many AIs are “still a little off or not easily distinguishable,” but added: “This isn’t about what GenAI can do today, it’s about the pace of change. Even if courts can barely keep up now, how will they react when fakes become indistinguishable from reality?”
” Mendones This is a warning shot. This shows the cost of infiltrating the system with AI counterfeiting. While the deepfakes in this case were crude enough for a judge to tell, technology has already advanced to the point that many of us have a hard time telling the difference. ”
Judge Schlegel, a member of the American Bar Association’s Special Committee on Law and AI, said the Federal Rules of Evidence Advisory Committee is considering rules that would require AI-generated evidence to meet the same reliability standards as expert testimony. However, he said that given that they are self-proclaimed litigants, they may not have been able to stop the incident.
