THE CAMERA DOES LIE: A NEW EVIDENTIARY BATTLEGROUND

For decades, the “gold standard” of defense evidence has been clear video footage. If a plaintiff claimed a slip-and-fall happened one way, but the security camera showed another, the case was often over. In 2025, that certainty is evaporating.

We are entering a new era of civil litigation where “seeing is believing” no longer applies. The rapid proliferation of generative AI has created two distinct risks for the defense: the introduction of fabricated deepfake evidence by plaintiff, and the strategic use of the “Liar’s Dividend” where opposing counsel dismisses authentic, incriminating video as “fake” simply because the technology to fake it exists.

This is no longer a theoretical risk. In September 2025, the Alameda County Superior Court issued a landmark ruling in Mendones v. Cushman & Wakefield, Inc. (Super. Ct. Alameda County, Sept. 9, 2025, No. 23CV028772). The court dismissed the plaintiffs’ case and issued terminating sanctions after determining that video exhibits submitted in support of a summary judgment motion were AI-generated deepfakes.

The videos, which purported to be witness testimony, featured robotic speech patterns and unnatural mouth movements that did not sync with the audio. While the court in Mendones caught the fraud because the fakes were “crude,” the judge noted that the technology is advancing so quickly that future fabrications may soon be indistinguishable from reality.

California lawmakers are racing to catch up, but the rules of the road remain unsettled. In October 2025, Governor Newsom vetoed Senate Bill 11, which would have mandated the Judicial Council to develop new rules for AI evidence admissibility. The Governor’s veto message suggested that existing laws were sufficient, effectively leaving defense counsel and trial judges to navigate this “wild west” without specific statutory guidance.

However, help is on the horizon. The California AI Transparency Act (SB 942) was signed into law and will fully take effect in 2026. This Act will eventually require large AI platforms to embed “provenance data,” or digital watermarks, into the metadata of AI-generated content.

The skepticism bred by deepfakes creates a secondary problem known as the “Liar’s Dividend.” As juries become aware that video can be faked, they may become hesitant to trust any video evidence. Plaintiffs may aggressively object to defense video evidence, such as dashcam footage or surveillance video, arguing it “could be” AI-generated, forcing defendants to spend time and money proving the authenticity of real evidence.

Recommended Strategies

To protect against these risks, we recommend that clients and counsel proactively adapt their discovery and evidence preservation protocols:

1. Litigants should move beyond accepting screen recordings or compressed video files. It is critical to specifically request “native file formats with intact metadata” in discovery. Deep-faking a video often strips or scrambles the original metadata, which can be the “smoking gun” of fabrication.
2. In the absence of a witness who can reliably say “that is exactly what happened,” a pristine, unbroken chain of custody from the recording device to the courtroom is the strongest defense against a deepfake allegation.
3. In high-exposure cases, counsel should consider engaging digital forensic experts earlier to analyze “pixel error” rates and lighting inconsistencies that human eyes might miss.

The Mendones sanctions prove that California courts are willing to punish AI fraud severely. The goal for attorneys is to ensure that when the evidence enters the courtroom, its authenticity is unassailable.