An Arizona appeals court has overturned a 10-year manslaughter sentence after the sentencing hearing included something no U.S. court had encountered before: an AI-generated video of the deceased victim, speaking directly to the judge. The case raises questions that go well beyond one courtroom in Chandler, Arizona. It touches on what counts as evidence, who gets to speak for the dead, and how far technology can go before it crosses a legal and ethical line.
How a Family Tried to Give a Victim His Voice
Christopher Pelkey was 37 years old when he was fatally shot during a road rage encounter at a stoplight in November 2021. He was unarmed. Gabriel Paul Horcasitas, who was convicted of manslaughter in Pelkey’s death, was sentenced to 10 years.
Before that sentence was handed down, Pelkey’s family did something unprecedented. His sister, Stacey Wales, had struggled to articulate what her brother would have wanted to say. Her solution was to use AI to reconstruct his likeness and give him a voice in the courtroom. The resulting video showed a digital version of Pelkey expressing a wish to still be with his loved ones, voicing a belief in forgiveness, and saying it was a shame he had crossed paths with Horcasitas because, in another life, they might have been friends. The video did not ask for any specific prison sentence.
Nine family members and friends had already spoken before the judge when the AI video was played. Wales has said her only intention was to humanize her brother for the court. The judge, Todd Lang, commented shortly before delivering the sentence that he “loved that AI,” though he did not state from the bench whether the video had influenced his decision. He did say that Pelkey’s “obvious forgiveness of Mr. Horcasitas reflects the character I heard about today.”
Why the Appeals Court Said No
The Arizona Court of Appeals, in a decision released in February 2026, ruled that Horcasitas must be resentenced. The three-judge panel concluded the AI video was not reliable. Its core objection was precise: the video did not document a real event or capture an actual moment. It presented a fabricated depiction of the victim’s thoughts, constructed from the imagination of his sister. The court found that presenting those invented words as if they came directly from Pelkey crossed a line that victim impact statements, however emotionally powerful, are not supposed to cross.
Horcasitas’ attorney had argued that her client had no meaningful opportunity to rebut the content of the video. She also noted that the video did not disclose who wrote the words attributed to the AI version of Pelkey, and that no evidence was presented to confirm the content accurately reflected his actual views. The “undue emotional weight” of seeing a deceased person appear to speak, she argued, was qualitatively different from a family member reading the same words aloud.
Prosecutors and the family’s legal team pushed back, arguing the video was an accurate representation of Pelkey’s character and that nothing in it was inflammatory. They also pointed out that Arizona law does not require victims to disclose in advance what they plan to say in court, and that Pelkey’s previous attorney had not objected to the video at the time. The Maricopa County Attorney’s Office stated that prosecutors knew the family would address the court but were not aware of the specific nature of what was planned.
What This Case Reveals About AI and the Law
Here is what most coverage of this story misses: the legal question is not really about AI. It is about authenticity and attribution. Courts have always had to decide what counts as a reliable representation of someone’s words or intentions. Wills get contested. Statements get challenged. The problem the AI video created was not that it used new technology. The problem was that it presented invented content as if it were the genuine expression of a real person who could no longer speak for himself or be cross-examined.
AI amplifies this problem in a specific way. A family member reading a prepared statement carries an implicit signal: this is one person’s interpretation of another person’s wishes. An AI-generated video of the deceased person, speaking in their own likeness, collapses that distance. It creates the impression of direct testimony where none exists. The emotional authenticity of the medium outpaces the factual authenticity of the content.
The case also signals that courts are entering territory where existing rules were not designed to go. AI use in the U.S. legal system has so far been concentrated in administrative tasks, legal research, and case preparation. Using it to generate victim impact statements is a different category entirely. Wales has said the family intends to continue finding ways to let Pelkey’s voice be heard through whatever means the court system allows. A victim appeals lawyer has already warned that using AI again could trigger another appeal.
In Short
An AI-generated video of a deceased victim, created by his sister and played at a sentencing hearing, led an Arizona appeals court to overturn a 10-year manslaughter sentence. The court’s objection was not to the emotion behind it, but to the fact that the video presented invented words as if they were the genuine voice of someone who could no longer speak. The case marks what is believed to be a first in U.S. courts, and it draws a boundary that will matter far beyond Arizona: AI can reconstruct a person’s appearance and voice, but it cannot reconstruct their actual testimony.
Based on reporting from Fast Company - Tech.