In the long human effort to let the dead speak in courts of law, Arizona has drawn a new line. A family's grief and love led them to reconstruct their murdered son and brother through artificial intelligence, giving him a voice of forgiveness at the sentencing of his killer — but an appeals court has ruled that what technology can simulate and what justice can accept are not the same thing. The conviction of Gabriel Horcasitas for the 2021 road rage killing of Christopher Pelkey stands, yet the sentence must be reconsidered, because the court found that a fabricated likeness of the dead, howev
Road rage killer's sentence overturned over AI victim video shown at trial
the AI video presents a depiction created from the imaginings of the victim's sister
So the conviction stands, but the sentence gets thrown out. That's an unusual split. What exactly did the appeals court say was wrong?
They said the video crossed a line because it wasn't a recording of something that actually happened. It was a reconstruction—the family used old recordings and photos to build an AI version of Christopher Pelkey, and then that AI version delivered a message of forgiveness to the man who killed him.
But hold on. The family didn't invent the forgiveness out of thin air, did they? Did Pelkey actually say those things before he died?
The reporting doesn't say that. It says they used voice recordings, videos, and pictures to recreate him. But it doesn't specify whether those recordings contained him actually saying those words.
That's a crucial gap. If he recorded a message of forgiveness before he died, that's very different from the family putting words in a dead man's mouth.
And the judge seemed to think it was genuine at the time. He said he heard forgiveness in it.
He did. But the appeals court said that's exactly the problem—the judge may have been influenced by something that wasn't actually the victim's voice or thoughts, but rather a construction by the family.
Which raises the question: would the sentence have been different without the video? We don't know that.
So this is really about drawing a line around what AI can do in court.
Exactly. The court is saying that even if the family's intentions were good, even if the message reflected what they believed Pelkey would have wanted to say, you can't use AI to put words in a dead person's mouth during sentencing.
And that precedent could matter a lot going forward, because victim impact is supposed to be about what actually happened, not what we imagine the victim might have wanted to say.
Il Polso
- A grieving family used AI to resurrect their loved one's voice and face, believing forgiveness deserved to be heard in the very room where his killer was judged.
- The sentencing judge was visibly moved by the video, raising immediate questions about whether an artificial reconstruction had quietly tilted the scales of punishment.
- Defense attorneys challenged the video as something the law had never sanctioned — not a recording of the victim, but a simulation of thoughts he never formally expressed.
- The Arizona Court of Appeals agreed, ruling that attributing words and feelings to a deceased person through AI crosses a line that victim impact law was never meant to permit.
- Horcasitas will face resentencing before a new judge, with the AI video removed from the record — one of the first explicit appellate rejections of AI-generated victim evidence in the United States.
In the long human effort to let the dead speak in courts of law, Arizona has drawn a new line. A family's grief and love led them to reconstruct their murdered son and brother through artificial intelligence, giving him a voice of forgiveness at the sentencing of his killer — but an appeals court has ruled that what technology can simulate and what justice can accept are not the same thing. The conviction of Gabriel Horcasitas for the 2021 road rage killing of Christopher Pelkey stands, yet the sentence must be reconsidered, because the court found that a fabricated likeness of the dead, however tenderly made, cannot be permitted to speak where only the living and the documented may.
Gabriel Horcasitas shot Christopher Pelkey dead at a red light in 2021, a road rage encounter that ended one man's life and began another's legal reckoning. A jury convicted him, and in 2024 a judge sentenced him to ten years — but the sentencing hearing contained something no American courtroom had quite seen before.
Pelkey's family, led by his sister Stacey Wales, had assembled an AI-generated video of the man they lost. Drawing on photographs, voice recordings, and video clips, they produced a reconstruction of Pelkey in a grey baseball cap, speaking in his own voice, delivering a message of forgiveness to the man who killed him. "In another life, we probably could have been friends," the AI version said. The judge thanked the family, acknowledged their pain, and indicated the video had shaped his understanding of the moment.
Horcasitas's attorneys appealed, and this week the Arizona Court of Appeals sided with them. The court drew a careful distinction: victim impact statements are permitted, and recordings made while a victim was alive are permitted — but this was neither. It was a simulation, attributing thoughts and words to someone who could no longer speak for himself, crafted from a family's intentions rather than the victim's own documented voice. That, the court ruled, crossed a line the law does not allow.
Horcasitas remains convicted. His sentence, however, is vacated, and a new judge will determine his punishment without the AI video in evidence. The case now stands as one of the first appellate rulings in the United States to explicitly reject AI-generated victim material — a boundary quietly but firmly drawn at the intersection of grief, technology, and justice.
Gabriel Horcasitas shot and killed Christopher Pelkey at a red light in 2021. It was a road rage confrontation. A jury found him guilty, and in 2024, a judge sentenced him to ten years in prison. But this week, an Arizona appeals court ordered him resentenced—not because the conviction was wrong, but because of what happened during the sentencing hearing itself.
Pelkey's family had created something unprecedented in that courtroom: an AI-generated video of their dead son and brother, speaking directly to the man who killed him. They built it from voice recordings, photographs, and video clips they had of Pelkey when he was alive. The AI reconstruction wore a grey baseball cap. It spoke in his voice. It delivered a message of forgiveness.
"To Gabriel Horcasitas, the man who shot me, it is a shame we encountered each other that day in those circumstances," the AI version said. "In another life, we probably could have been friends." The video continued: "I believe in forgiveness, and a God who forgives. I always have and I still do."
At the time, Judge Todd Lang seemed moved by it. He thanked the family for the video, acknowledged the family's anger as justified, and said he heard genuine forgiveness in what had been presented. The judge's words suggested the AI message had shaped how he understood the case at the moment of sentencing.
Horcasitas's legal team appealed, arguing that allowing the video was a mistake. This week, the Arizona Court of Appeals agreed. The court wrote that the AI video "crossed that line"—it did not document something that actually happened or preserve a moment that was recorded. Instead, it created a depiction of the victim and attributed thoughts to him that came from the imaginings of his sister, Stacey Wales, who had overseen the video's creation.
The distinction matters legally. Victim impact statements are allowed in sentencing. A video of the actual victim, recorded while alive, would be allowed. But this was neither. It was a simulation of a person who could no longer speak, designed to convey a message the family wanted conveyed, using technology to make it seem as though the dead man himself was speaking. The appeals court found that this crossed into territory that should not be permitted in a courtroom.
Horcasitas remains convicted of the killing. His ten-year sentence is not final, though. He will be resentenced, and a new judge will decide his punishment without the AI video in the record. The case marks one of the first times a U.S. court has explicitly rejected AI-generated victim evidence on appeal, setting a boundary around what technology can do in the pursuit of justice.
Citazioni salienti
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— Arizona Court of Appeals
I believe in forgiveness, and a God who forgives. I always have and I still do.— AI-generated message attributed to Christopher Pelkey