A court has ruled on evidence that speaks in a dead man’s voice
The Arizona Court of Appeals has thrown out a 10.5-year manslaughter sentence because the judge who imposed it had watched an AI-generated video of the victim addressing the court. The panel left the conviction standing and sent the case back for resentencing.
The unanimous, published opinion in State v. Horcasitas came down on 30 September in Division One. It is the first American appellate decision to rule squarely on whether a synthetic recreation of a deceased victim can be used at sentencing.
Gabriel Paul Horcasitas was convicted of manslaughter in the 2021 road-rage shooting of Christopher Pelkey in Chandler. At sentencing, Pelkey’s sister Stacey Wales presented a video that combined real footage of her brother with an AI reconstruction built from his photograph and voice, which appeared to show him speaking directly to the courtroom, to his family and to the defendant, and offering forgiveness.
Why disclosure was not enough
Everyone in the courtroom knew the video was synthetic. The appeals court held that this did not fix the problem.
The panel’s reasoning turns on whose views the video actually carried. The family, it wrote, “presented his sister’s conception of what he would have thought and said through Pelkey’s recreated face and voice, eliminating the distance between the family’s understanding of his views and Pelkey’s actual views”. A statement about the victim becomes, once rendered in his face and voice, a statement by him — and he cannot confirm it.
On that basis the court found the recreation lacked the reliability required of information used to determine a criminal sentence, and that its use made the sentencing fundamentally unfair.
The trial judge’s own words are in the opinion. Judge Todd Lang said he “loved that AI”, described the recreation as genuine, and referred to Pelkey’s apparent forgiveness as coming from his heart. Horcasitas received 10.5 years, the presumptive maximum; prosecutors had asked for nine.
What this does and does not decide
The ruling is narrow in one direction and broad in another. It does not touch the conviction, and it does not bar victim impact statements that describe what a victim would have wanted — families have always been allowed to say that in their own voices. What it bars is putting those words into the victim’s reconstructed mouth.
That distinction will travel. Sentencing is the stage of a criminal case with the loosest evidentiary rules: judges can consider material that would never reach a jury. Arizona has now held that the looseness has a floor, and that a synthetic likeness falls below it.
The practical effect is immediate for any court that has allowed these videos, and several US jurisdictions have. Generating one requires a photograph, a short voice sample and consumer software. The Pelkey video was reported widely when it was first shown, largely as a story about a grieving family’s ingenuity. The appellate record reframes it as a question about what a judge was actually looking at.
What happens next
Horcasitas returns to Maricopa County Superior Court for resentencing before a judge who will not see the video. The conviction stands, so the only question is the term.
The wider thing to watch is whether other state appellate courts follow Arizona or split from it. A split would eventually put the question in front of a federal court, and the answer would then apply to a practice that currently has no rule at all.
