Arizona Appeals Court Vacates Road-Rage Sentence After AI Video Spoke for Slain Victim

Gabriel Paul Horcasitas must be resentenced after the Arizona Court of Appeals vacated his sentence in the 2021 road-rage shooting death of Christopher Pelkey. Horcasitas was convicted of manslaughter and received a prison term of about 10 years, with reports placing the sentence at 10 years or 10 1/2 years. A three-judge panel ruled that an AI-generated video shown before sentencing created a fundamental error that made the punishment unfair.

The appeals court found that the video affected the sentencing judge, who referred to the forgiveness expressed in the digital statement. The ruling leaves Horcasitas’ conviction in place while requiring a new sentencing proceeding. The case places a new boundary around how digitally reconstructed people can be used in criminal court.

Pelkey’s family created the video after his death using a prior recording, a photograph and a script written by his sister, Stacey Wales. The resulting digital depiction showed Pelkey addressing the courtroom and Horcasitas before the sentence was imposed. The video conveyed forgiveness, giving the sentencing judge a message presented through Pelkey’s likeness and simulated voice.

The family’s purpose was to give Pelkey a voice after his November 2021 death during a confrontation at a red light in Chandler, Arizona. The legal dispute centered on how that message was presented, rather than on whether Pelkey’s relatives were entitled to speak about their loss. The appeals panel distinguished a recording of an actual event from an AI depiction built from another person’s interpretation of the victim’s thoughts.

The court said the video did not document an event or capture a statement Pelkey made while alive. Instead, it portrayed the victim and thoughts created from the imaginings of his sister, according to the ruling. That distinction mattered because the video presented words as though they came directly from Pelkey, while the script originated with a surviving family member.

The panel described the AI presentation as unreliable and said it crossed the line into a form of evidence that prejudiced the sentencing decision. The judge’s reference to Pelkey’s forgiveness was central to the finding that the video had a concrete influence. Reports characterize the decision as a ruling about sentencing fairness, not a determination that AI tools are categorically barred from every courtroom use.

The new proceeding will determine Horcasitas’ punishment without the AI victim statement that the appeals court found improper. His manslaughter conviction remains intact, so the ruling does not order a new trial or erase the jury’s finding that he killed Pelkey. The next hearing will instead revisit the sentence under procedures that exclude the disputed digital depiction.

The case also gives judges and lawyers a practical warning: a digitally recreated victim can carry emotional authority that ordinary family testimony does not. A judge may hear a relative describe what the deceased believed, but an animated likeness speaking in the victim’s apparent voice can make an interpretation feel like direct testimony. That difference now has direct consequences in an Arizona sentencing case.

The ruling arrives as courts confront synthetic voices, faces and videos that can imitate people who never recorded the words attributed to them. In this case, the technology was used by a victim’s family rather than by a commercial litigant, but the legal issue is broader: whether an AI reconstruction gives a fictionalized statement the persuasive force of firsthand evidence.

The decision also separates emotional expression from evidentiary reliability. Family members can present grief, forgiveness and views about punishment, yet the court found that those views cannot be made to appear as an authentic posthumous statement from the victim. Lawyers handling future sentencing hearings will have to identify who wrote the words, what source material supports them and whether a digital presentation changes how the judge receives the message.

Horcasitas will return for resentencing, while the Arizona appellate decision remains a notable precedent on AI-generated victim-impact material. The immediate legal question is the length and terms of a new sentence, not Horcasitas’ guilt. The case also leaves prosecutors, defense lawyers and judges with a clear procedural concern: emotionally powerful synthetic media can affect punishment even when it contains no recording of an actual statement.

For families, the ruling does not eliminate the ability to speak about a victim or seek accountability. It draws a line between family testimony and a generated likeness that appears to speak for the dead. As courts encounter more AI-created evidence, authentication, authorship and the risk of emotional distortion will become central parts of sentencing disputes.