NATIONAL CENTER FOR STATE COURTS RAISES ALARM OVER AI-GENERATED EVIDENCE

A recent article by the National Center for State Courts warns about the threat AI-generated evidence poses to “public trust in the courts.”

The article cites a California civil case from September 2025, as likely the first instance “where a deepfake was submitted as purportedly authentic evidence and detected as AI-generated.”

The plaintiffs in the case were self-represented, and the submitted video was crude and easily identified as a deepfake, leading Alameda County Superior Court Judge Victoria Kolakowski to issue a “termination” order dismissing the case, the report notes.

In another instance, a Florida woman spent two days in jail when her ex-boyfriend allegedly used AI to fabricate a string of fake texts, leading to her arrest for violating a protective order, the report notes.

Prosecutors later dropped the charges.

To date, there are no confirmed cases of lawyers knowingly submitting AI-generated deep fakes, the NCSC reports, “but they are far from a perfect bulwark against dubious AI-generated content entering the courtroom.”

Judges across the country began voicing concerns last year in national news stories that also quoted NCSC experts.

“I think AI-generated fake or modified evidence is happening much more frequently than is reported publicly,” Santa Clara County Superior Court Judge Erica Yew told NBC News.

Yew was identified as an ardent supporter of using AI to increase court access. However, she acknowledged that she is concerned that it would be difficult to detect a deep-fake audio recording that was submitted to support a protective order. She also expressed concern that AI technology could be used to fake legal instruments that the courts take at face value, such as the title to an automobile.

“We’re in a whole new frontier,” she said.

In 2024, the NCSC partnered with the Tomson Reuters Institute to form the “AI Policy Consortium for Law & Courts.” A consortium policy paper “Evidentiary Issues Raised by Artificial Intelligence” points to another challenge associated with AI-generated evidence.

“The mere existence of AI technology casts doubts on the authenticity of all digital evidence, potentially undermining the foundation of fact-finding in legal proceedings,” the paper states. “This potentially creates a new paradigm where the authenticity of digital evidence is constantly in question.”

The phenomenon is referred to as the “Liar’s Dividend” – the potential for a guilty party to benefit from doubt cast on authentic evidence.

The U.S. Judicial Conference’s Advisory Committee on Evidence Rules last year considered two proposed rules to address the challenges, including one by Duke Law School Professor Paul Grimm, a former federal judge, and Maura R. Grossman, a research professor of computer science at the University of Waterloo. It would require parties that allege the opposition used deepfakes to thoroughly substantiate their arguments. The committee failed to approve the proposals, with the majority arguing that existing standards of authenticity are “up to the task.”

The AI consortium is offering a library of more than two dozen webinars and other resources to help judges and lawyers address challenges associated with AI-generated evidence.

A bench card refers to “Unacknowledged AI-Generated Evidence” to distinguish deep fakes from evidence that both sides acknowledge was generated by AI, such as an accident reconstruction video.

The bench card suggests questions for the proffering party, including “Where any forensic tools or methods used to verify the integrity of the evidence?” Another asks, “Can you provide metadata or other technical information that supports the authenticity of this digital file?”

A “Judicial Actions” section suggests “If the proffering party’s answers to the questions above are incomplete or raise concerns, then request clarification, supporting documents and files, or expert testimony.”