In the first installment of our blog series, Artificial Intelligence or Artificial Interference?: How AI is Reshaping Litigation for Better and Worse, we wrote about how artificial intelligence (AI) is beginning to have an impact on litigation and various pitfalls created by reliance on AI in the legal context. Late last year, a noteworthy case brought those concerns to fruition after two plaintiffs introduced AI-generated and materially altered exhibits as “evidence.” To be sure, artificial intelligence has its benefits, but, as Mendones v. Cushman and Wakefield, Inc. shows, AI presents significant risks, and attorneys and litigants alike must be cognizant of the impact AI is having throughout the litigation process.
Background
In Mendones, a case arising out of California state court, Ariel and Maridol Mendones (“Plaintiffs”) were accused of submitting intentionally false testimony in connection with their motion for summary judgment. Specifically, the court suspected that Plaintiffs submitted 9 different exhibits that were either AI-generated or altered. The court highlighted particular examples. For instance, Plaintiffs submitted a video testimony that purported to be a recording of a woman by the name of Geri Haas. The court concluded that “[t]he accent, cadence, volume, word choice, pauses, gestures, and facial expression, among other characteristics, of the person depicted in [the real exhibit] are vastly different from those demonstrated by the ‘person’ depicted in [the generative AI exhibits]. The ‘person’ depicted in [the generative AI exhibits] lack expressiveness, are monotone, do not pause at moments where pauses are expected, use odd words [sic] choices, and appear generally robotic. Further, the mouth flap does not match the words being spoken.”
Plaintiffs also submitted text messages that the court found to be products of generative AI, or at the very least, materially altered. In fact, it was so obvious to the court that the messages were fabricated, that Plaintiffs’ declaration about the veracity of these messages “strain[ed] credulity.”
After going through the above examples and more, the court noted that it “remain[ed] suspicious of the other evidentiary submissions,” but that it did not “have the time, funding, or technical expertise to determine the authenticity of Plaintiffs’ statements or conduct a forensic analysis of the suspect evidentiary submissions.”
The court then finally turned to the appropriate sanction, where it considered three options: (1) monetary penalties, (2) criminal referral, and (3) termination of litigation. Monetary sanctions and referral for criminal prosecution were found to not be appropriate, as the former was too light, and the latter too severe. This resulted in a terminating sanction, i.e., dismissal of the lawsuit, which the court said was “proportional to the harm that Plaintiffs’ misuse of the Court’s processes ha[d] caused.” In its final few remarks, the court made the following clear: it has zero tolerance for AI “evidence.”
Not-So-Futuristic Applications
We previously mentioned the possibility that a litigant could offer an AI-generated video of a witness’s deposition testimony, noting that AI-generated video or audio, complete with synthesized voice, tone, and body language, adds a new layer of complexity and risk to the court and to litigants. Mendones makes that statement almost prophetic. While AI in its current state may not fool a court or opposing counsel, its mere introduction into an actual litigation matter is concerning, particularly because the Mendones court recognized that it did not have the resources to analyze all potentially fabricated exhibits. With increasingly burdened dockets, tight timelines, and pressing deadlines, many courts may be in the same boat. As AI develops, the risk of fabricated evidence being submitted, or even admitted, heightens.
There is no doubt that AI will be used in the court room again, despite all the warnings judges and attorneys alike put forth. As with any new technology, people will push the boundaries to see what they can do and/or get away with. While the Mendones court refused to admit such “evidence,” another court may come to a different conclusion or fail to recognize the AI-generated or -manipulated nature of evidence entirely. Therein lies the risk of allowing AI-generated witness testimony or AI-enhanced evidence in litigation. The ability to use AI to create evidence that does not actually exist moves across the line from artificial intelligence to artificial interference with the opponent’s right to a fair trial.
Key Takeaways for Litigants
- Use AI at your own risk. In our previous blog, we mentioned how AI is still a very new technology. While it may, in some circumstances, streamline time-consuming research, writing, or discovery projects or allow individuals to organize their thoughts or synthesize data in a coherent way, using it to fabricate evidence could cost a litigant far more than the attorneys’ fees saved by using AI as a shortcut. Additionally, represented litigants should request that their attorneys disclose the use of AI tools in litigation. The improper use of AI can lead to significant penalties, and litigants should know those risks when engaging counsel.
- Be ready for AI “evidence” in litigation. Between the Horcasitas case, mentioned in the first installment of this series, and Mendones, it is becoming increasingly more likely that litigant opponents are going to rely on AI to pursue whatever perceived advantage they imagine they can gain with it. As Mendones demonstrates, this evidence could be presented to a court at a hearing or trial. But AI-generated evidence could impact litigation in other ways as well. For instance, a party could produce AI-generated or altered evidence during the discovery process to create a “smoking gun” document or video that materially alters the opposing party’s risk analysis. A litigant could choose to settle, or increase a monetary settlement authority based on evidence completely fabricated by the opponent. So, we must reiterate that when AI-generated or altered evidence is offered at any stage of the litigation, the opposing party must be prepared to vet that evidence, including by inquiring about the method and manner of its creation, the person requesting or participating in its creation, whether there were any other outputs generated, what changes were made to prompts to lead to the final result, and whether the evidence was reviewed and approved by a qualified third party expert. If evidence seems too good or too bad to be true, that just may be the case. Only where a litigant is fully appraised of the source and content of all aspects of an opposing party’s case, from pre-suit demand evidence all the way through trial, can that litigant present their best position and obtain the fair proceeding guaranteed by our justice system.
Should you have questions regarding AI use in litigation, please contact Matthew Feinberg, Adel Mansour, or another member of PilieroMazza’s Litigation & Dispute Resolution practice group.
Related Attorneys
