
Arizona AI Victim Video Leads Court to Vacate Sentence
Published by AINave Editorial
An Arizona appeals court has vacated Gabriel Horcasitas’s 10½-year prison sentence after the sentencing judge allowed and relied on an AI-generated video depicting the man he killed, Christopher Pelkey, forgiving him. The panel reaffirmed Horcasitas’s manslaughter conviction and ordered a new sentencing, so this ruling changes the punishment proceeding, not the verdict. The court’s decision was issued September 30.
The video mattered at sentencing, not at trial
The case stems from Pelkey’s 2021 death in a road-rage shooting. Horcasitas, 55, was convicted of manslaughter and had received a 10½-year sentence. The three-judge panel said the sentencing judge committed a “fundamental error” by allowing and relying on a video that generated a likeness of Pelkey’s voice and appearance. The sentence was thrown out while the conviction was reaffirmed.
That distinction is important. The ruling does not undo the finding that Horcasitas was guilty, nor does the supplied account describe a new trial. It requires the court to determine his sentence again without the sentencing decision resting on the video in the way the appeals panel criticized.
A generated likeness can carry a human message
The video depicted Pelkey forgiving Horcasitas. That makes the issue more than the use of synthetic voice or imagery: the generated portrayal put words and a stance into the apparent voice of someone who could not speak for himself at the hearing. The appeals panel focused on the judge’s decision to allow and rely on that AI-generated likeness.
In a sentencing hearing, a victim’s apparent words may have emotional weight. When AI supplies the appearance and voice, the court must confront what authority those generated words should have in a consequential decision. That is a reasoned implication of this case, not a rule the available report establishes for every courtroom use of AI.
What the decision leaves open
Horcasitas must be resentenced, but the report gives no date for that proceeding. It also does not set out a broader test for AI-generated material in court. The safe reading is narrow: the Arizona appeals court found fundamental error in this sentencing, where the judge both allowed and relied on the video. The manslaughter conviction remains affirmed.
The case therefore draws a practical line between a generated likeness being presented and that likeness influencing punishment. The conviction stands; the sentence must be decided again.






















