Arizona court orders resentencing after AI video of dead victim played in court

An Arizona appeals court has thrown out the prison sentence of a man convicted of a road rage killing because a video generated with artificial intelligence, in which his dead victim appeared to address the court, was played at the sentencing hearing. The guilty verdict stands, but the case must now return to court for a new sentencing, according to the BBC.

The killing and the original sentence

According to the BBC, Gabriel Paul Horcasitas, 55, was found guilty by a jury of shooting and killing 37-year-old Christopher Pelkey during a confrontation at a red light in 2021. He was sentenced to 10 years in prison.

Ahead of the sentencing hearing in 2025, Pelkey's family played a video in which an AI reconstruction of Pelkey delivered what was presented as a victim impact statement. His sister, Stacey Wales, told the BBC last year that the family had assembled the recreation from voice recordings, videos and photographs of him.

What the AI version of the victim said

The generated likeness of Pelkey, shown wearing a grey baseball cap, spoke in the first person and addressed the man who killed him. It said it was a shame the two had crossed paths that day, that "in another life, we probably could have been friends," and that it believed in forgiveness.

The judge who oversaw the sentencing, Todd Lang, embraced the technology at the time. "I loved that AI, thank you for that," he said from the bench, adding that the forgiveness came across as genuine.

Horcasitas's lawyer, Kristen Reller, appealed on the argument that the trial judge should never have allowed the video into the hearing. Reller declined to comment after the ruling, and an attorney for victims in the case, Jessica Gattuso, did not respond to a request for comment, the BBC reported.

What the appeals court decided

The Arizona Court of Appeals sided with the defence, ruling that the AI clip "crossed that line" and that Horcasitas must be resentenced.

In its written decision issued on Wednesday, the court drew a sharp distinction between real evidence and synthetic testimony. Rather than documenting an actual event or preserving something the victim actually said, the court wrote, the video amounted to "a depiction of the victim and his thoughts created from the imaginings of the victim's sister."

In other words, the appeals court treated the video not as a record of Pelkey but as a persuasive artifact produced by someone else and voiced through him, which the sentencing judge then credited as though it were the victim's own sentiment.

Why it matters

The ruling is a significant precedent for how courts handle generative AI, and it lands at a moment when the technology can already produce convincing video and voice recreations from modest source material.

The case illustrates the specific danger appellate judges identified: sentencing hearings are meant to weigh real testimony and real harm, and an AI rendering can inject emotionally powerful words into the mouth of someone who cannot confirm, deny or author them. That the trial judge found the generated forgiveness "genuine" is exactly why the higher court stepped in, since the simulated sentiment plausibly influenced the outcome.

Notably, the conviction itself was untouched. The appeals court's concern was confined to the sentencing phase, where victim impact statements carry weight, which signals to lower courts that even sympathetic, family-made AI content is off limits there.

For courts beyond Arizona, the decision offers an early appellate blueprint: AI-generated depictions of people, living or dead, do not simply slip into existing categories of evidence. As tools for recreating voices and faces become commonplace, judges will need explicit rules for what synthetic media may be shown, and this case now stands as the reference point for saying no.