An AI Video of a Murdered Man Helped Send His Killer to Prison for 10.5 Years. Arizona Just Threw That Sentence Out

An Arizona man will be resentenced because a judge watched an AI-generated video of the man he killed "speak" in court. The Arizona Court of Appeals has vacated the 10.5-year manslaughter sentence of Gabriel Paul Horcasitas, ruling that an AI rendering of victim Christopher Pelkey was unreliable and so prejudiced the sentencing judge that the proceeding became fundamentally unfair. The conviction stands; the punishment does not.

The underlying case is a 2021 road-rage shooting at a stoplight in Chandler, a Phoenix suburb, in which Horcasitas, now 55, fatally shot the 37-year-old Pelkey, an unarmed US army veteran who had stepped out of his truck. In May 2025, at the sentencing hearing, Pelkey's family played a roughly four-minute video in which an AI likeness of him — built by his sister Stacey Wales from a photograph, voice recordings and a script she wrote — addressed his killer directly.

"It is a shame we encountered each other that day in those circumstances," the digital Pelkey said. "In another life, we probably could have been friends. I believe in forgiveness, and a God who forgives. I always have and I still do." Judge Todd Lang said on the record that he "loved that AI," and then imposed 10.5 years — a year more than prosecutors had requested.

The three-judge appeals panel did not ban synthetic victim statements. It found something arguably more consequential: that no framework exists for admitting one. There is no authentication standard, no foundation requirement, and no way for a defense to cross-examine a dead man rendered at 24 frames per second. "Rather than document an event or recording a particular moment," the court wrote, "the AI video presents a depiction of the victim and his thoughts created from the imaginings of the victim's sister."

That framing identifies the actual harm. A family member reading a statement aloud is transparently an interpretation — a court can weigh it as grief, not testimony. An AI avatar erases that distance. As the panel put it, the video "erases the interpretive distance between the family's belief of what the victim would have said and the victim's own voice and opinions," conveying an authenticity that the same words, spoken by anyone else, would not have carried.

The family's position is not unreasonable, and the court did not treat it as bad faith. Wales said her only goal was to humanize her brother for the judge, and she now argues the system protects the convicted more than the bereaved: "There is a convicted murderer sitting in prison that has blankets and walls of protection around their rights. Where are the rights for the victim?" Prosecutors at the Maricopa County Attorney's Office said they knew the family would address the court but were not aware of the nature of the statement.

AI in the courtroom has so far stayed in the administrative lane — legal research, case preparation, and in Arizona, helping explain rulings to the public. Victim impact statements were the natural next frontier precisely because they sit outside the evidentiary phases of a trial, where guardrails are deliberately loose. That looseness is exactly what made a photorealistic reconstruction so potent, and so dangerous, in a sentencing hearing.

Expect the practical consequences before the legislative ones. Defense attorneys now have a template appellate ruling to cite wherever an AI avatar has appeared in a victim statement, and several US states have quietly allowed such reconstructions over the past 18 months. Courts will need authentication standards, disclosure rules, and an answer to the basic question this case raises: whether the words a synthetic victim speaks are the victim's voice at all, or a family's grief rendered in someone else's likeness.