Arizona Court Rules Victim Impact Evidence Created with Artificial Intelligence Inadmissible

It is virtually impossible to read or watch the news these days without encountering advances in artificial intelligence. As the technology advances, lawyers have been finding creative ways to use the technology to advance their respective positions. Recently, AI-generated victim impact evidence was presented during a criminal sentencing. As explained below, the Arizona Court of Appeals concluded that the evidence was unreliable and ordered a new sentencing.

 

• *“Victim Impact Evidence” is testimony or statements from victims, their family, friends and/or loved ones describing how the criminal act(s) at issue have affected them personally. Victim Impact Evidence is routinely offered during sentencing and can carry a lot of weight with judges when deciding on the appropriate sentence to impose.

 

In a recent manslaughter case out of Arizona,  an AI-generated video produced by the victim’s sister was played for the court during sentencing. The video recreated the deceased victim's likeness and voice profile. The AI "victim" spoke directly to the court, family, and the defendant, stating: "this here is a true representation of who I was . . . not how the courtroom portrayed me" and expressed forgiveness. In imposing the 10.5-year presumptive term, the sentencing judge explicitly stated: "I loved that AI. Thank you for that... I feel that that was genuine; that his obvious forgiveness of [the defendant] reflects the character I heard about today..." On appeal, the appellant argued that the court erred by considering this evidence.

 

The Arizona Court of Appeals applied  fundamental error review, noting that while the Arizona Victims' Bill of Rights affords victims broad rights to be heard at sentencing, evidence must bear minimal indicia of reliability to satisfy constitutional Due Process. Sentences based on inaccurate or unreliable information must be set aside. The Court concluded that the AI video did not document a real event; it presented statements and non-verbal expressions created from the sister's subjective imaginings of what the victim would say. It erased the interpretive distance between the family's beliefs and the actual victim. The Court concluded that the presentation crossed the line into undue prejudice. Because the record demonstrated that the sentencing judge expressly relied on the AI video as "genuine" and factored it into reducing the weight of Horcasitas's mitigation evidence, the sentencing proceeding was fundamentally unfair.

The Court reversed and remanded for resentencing.

 

As AI is being adopted by lawyers and used in their practices, new rules about admissibility and reliability will be needed. The manner in which the defendant’s rights were violated by the use of AI in this case underscores the need for such legal guideposts.

 

The Arizona Court of Appeals decision is available here: State v. Horcasitas - 1 CA-CR 25-0191 - Opinion.pdf