Arizona Court Overturns Manslaughter Sentence Over AI-Generated Victim Impact Statement

In a landmark legal development that underscores the complex intersection of emerging technology and the American judicial system, the Arizona Court of Appeals has vacated a 10-year manslaughter sentence for Gabriel Paul Horcasitas. The decision, handed down this past Wednesday, centers on a controversial evidentiary maneuver: the use of an artificial intelligence-generated video in which the deceased victim, Christopher Pelkey, appeared to speak directly to the judge during the sentencing phase.

The three-judge appellate panel concluded that the reliance on the AI-generated footage was legally flawed, noting that the video was inherently unreliable and failed to reflect objective reality. By presenting a digital simulation that purported to articulate the thoughts and sentiments of the deceased, the court determined that the process had veered into the realm of conjecture rather than evidence, necessitating a resentencing for Horcasitas.

The underlying case dates back to a November 2021 road rage incident in Chandler, Arizona, a suburb of Phoenix. According to authorities, the encounter at a local stoplight turned fatal when 55-year-old Gabriel Paul Horcasitas shot and killed 37-year-old Christopher Pelkey. At the time of the shooting, Pelkey was unarmed; he had exited his truck and was walking toward Horcasitas’s vehicle when the confrontation escalated. Horcasitas was subsequently convicted of manslaughter and endangerment, the latter charge stemming from a stray bullet that struck a nearby vehicle during the encounter.

The sentencing hearing in May 2025 became the stage for a technological experiment that many legal experts believe is a first in the United States court system. Before the judge delivered the sentence, Pelkey’s family sought to provide the deceased with a final voice. His sister, Stacey Wales, spearheaded the effort, using AI software to create a digital likeness of her brother. Wales later explained that she had struggled to articulate her brother’s perspective and character, leading her to believe that an AI rendering could bridge that emotional gap for the court.

The resulting video, which featured a synthetic version of Pelkey, offered reflections on his life and death. The digital avatar expressed a desire to remain with his loved ones, articulated a belief in the necessity of forgiveness, and even suggested that he and his killer might have been friends “in another life.” The video did not explicitly lobby for a specific prison sentence, but its presence in the courtroom created a profound, if legally dubious, impact.

However, the appellate court was unmoved by the emotional intent of the family. “Indeed, rather than document an event or recording a particular moment, the AI video presents a depiction of the victim and his thoughts created from the imaginings of the victim’s sister,” the panel wrote in their ruling. The court found that the video crossed a fundamental threshold by presenting a manufactured narrative as if it were the authentic, verified voice of the deceased.

Defense attorney Kristen Reller, who represented Horcasitas, had consistently argued that the trial judge, Superior Court Judge Todd Lang, violated due process protections by allowing the AI evidence to influence the proceedings. Reller contended that the video carried an “undue emotional weight,” conveying a sense of authenticity that would have been impossible to achieve had a human family member simply read those words aloud. Furthermore, Reller pointed out that the defense had no meaningful opportunity to cross-examine or rebut the material presented by the AI, as the origins of the script and the parameters of the AI’s generation were never fully disclosed.

The prosecution and victim’s legal representatives had defended the use of the video, arguing that it served as an accurate reflection of Pelkey’s character. They maintained that the judge did not err, as the footage was merely a medium for victim impact testimony, which is protected under Arizona law. Victims are afforded the right to speak before the court, or to submit their statements via written, audio, or video formats. The family’s legal team argued that the judge was already well-informed by the live, emotional testimony of nine friends and family members who had testified earlier in the hearing, and that the AI video was merely a supplement to that testimony.

In court records, the lawyers for the victim’s family further argued that Arizona law does not mandate the disclosure of victim impact statements to prosecutors or defense counsel in advance. They also noted that the defense did not initially object to the video when it was presented.

The trial judge, Todd Lang, appeared to view the footage with interest at the time. Shortly before delivering the sentence, he remarked that he “loved that AI,” though he did not explicitly state on the record whether the video was a decisive factor in his sentencing calculations. He did, however, comment that Pelkey’s “obvious forgiveness of Mr. Horcasitas reflects the character I heard about today.”

The fallout from the appellate court’s decision has left the victim’s family in a state of frustration. Stacey Wales expressed her disappointment, emphasizing that her sole intention was to humanize her brother for the court. “It feels unfair because there is a convicted murderer sitting in prison that has blankets and walls of protection around their rights,” Wales said. “Where are the rights for the victim?”

When asked about the potential for further legal challenges, Wales indicated that the family remained undeterred. She noted that a victim appeals lawyer had cautioned her that using AI again could trigger another appeal, yet she remained resolute. “We will continue to let his voice be heard in whatever allowable medium we can convey that through the court system,” she said.

The case has ignited a broader conversation about the limits of technology in the courtroom. While artificial intelligence is increasingly integrated into the legal sector, its application has largely been confined to administrative tasks, such as legal research, data management, and public communication. In Arizona, for instance, AI has been utilized to summarize and inform the public about rulings in high-profile cases. However, the use of AI to generate synthetic personas—particularly those of deceased individuals—represents a significant departure from standard practice.

Legal scholars suggest that this case serves as a warning for future proceedings. As AI tools become more sophisticated and accessible, the temptation to use them to enhance emotional narratives in court will likely increase. Yet, as the Arizona Court of Appeals ruling demonstrates, the judicial system relies on the integrity of evidence. When that integrity is compromised by synthetic, unverified, or manufactured content, the fairness of the judicial process is called into question.

The Maricopa County Attorney’s Office, in a statement released Thursday, confirmed that while prosecutors were aware the victim’s family would be providing statements during the sentencing, they were not briefed on the nature of the AI-generated video. The lack of clear guidelines regarding the use of such technology in victim impact statements has left both the prosecution and the defense navigating uncharted territory.

As the case returns to the lower court for resentencing, the decision serves as a definitive signal that the "AI era" in the courtroom will be subject to strict scrutiny. The appellate court’s ruling establishes that, regardless of the emotional value or the desire for closure, the fundamental requirements of due process and the reliability of evidence remain the bedrock of the American legal system. The case of Gabriel Paul Horcasitas will likely be cited in future debates regarding the boundaries of victim rights, the potential for AI to introduce bias, and the necessity of ensuring that the pursuit of justice does not rely on the artificial simulation of the truth.