An Arizona appeals court just vacated a sentence because of an AI-generated victim video
On September 30, 2026, the Arizona Court of Appeals held that a sentencing judge who relied on an AI-generated video of a homicide victim made the sentencing fundamentally unfair. The court said no Arizona case had addressed an AI-generated depiction of a victim as victim impact evidence. The decision answers a question every sentencing lawyer is now going to be asked: how reliable does something have to be before a judge may rely on it?
By John A. Julian, Esq. Published October 1, 2026.
Key takeaways. An AI video of a deceased victim, scripted by a family member, was played at sentencing. The judge said he "loved" it and found it "genuine." The appeals court held that this was fundamental error, vacated the sentence, and ordered resentencing, while leaving the conviction in place. The ruling is Arizona law and does not bind Pennsylvania courts, but the due process principle behind it applies everywhere: a sentence has to rest on reliable information.
What happened
Gabriel Paul Horcasitas shot and killed Christopher Pelkey during a road rage confrontation at a Chandler, Arizona intersection in November 2021. A jury rejected his self-defense claim and convicted him of manslaughter. At sentencing in May 2025, the victim's family gave statements, and the victim's sister then played a video in which an AI-generated version of the victim, built from his photograph and a voice profile, spoke to the court and to Horcasitas. The avatar said the video was "a true representation of who I was," told Horcasitas "I believe in forgiveness and in God who forgives," and thanked the people in the courtroom. Parts of the video were real footage of the victim, which the appeals court found permissible. The rest was scripted by his sister.
The sentencing judge responded on the record: "I loved that AI." He said he felt the video "was genuine" and that the victim's "obvious forgiveness" of Horcasitas reflected the character he had heard about. He then imposed 10.5 years, the presumptive term, which was more than the 9 years the prosecutor had requested. The family had asked for the maximum.
What the appeals court held
In State v. Horcasitas, No. 1 CA-CR 25-0191 (Ariz. Ct. App. Sept. 30, 2026), a three-judge panel affirmed the manslaughter conviction and vacated the sentence. Its reasoning, in the court's own terms:
- Victims have a right to be heard, but it is not unlimited. Arizona's Victims' Bill of Rights lets victims address the court at sentencing without taking an oath or facing cross-examination, and those rights are construed liberally. They still cannot infringe the defendant's right to due process.
- Sentencing evidence must be reliable. The usual evidence rules relax at sentencing, but only information bearing "minimal indicia of reliability" may be considered. A defendant has a due process right to be sentenced on accurate information.
- The AI video was not reliable. Apart from the embedded real footage, the video "does not reflect actual events" and instead presents "a depiction of the victim and his thoughts created from the imaginings of the victim's sister." It "erases the interpretive distance between the family's belief about what the victim would have said and the victim's own voice and opinions." The court added: "No disclaimer could cure that error."
- It mattered to the judge. "While the record does not reflect precisely how the AI video factored into the sentencing calculus, there is no question it played a role."
The court noted that no Arizona case had addressed an AI-generated depiction of a victim as victim impact evidence. It relied on the state supreme court's decision in State v. Rose, which warned prosecutors and victims "not to venture too close to the line" and encouraged judges to screen victim impact presentations.
Three details that matter beyond the headline
The defense never objected, and still won
Horcasitas's lawyer did not object to the video at sentencing, so the court reviewed only for fundamental error. It found that standard met. That is useful to know, but it should not be read as permission to stay quiet. Fundamental-error review is a hard standard, and the court leaned heavily on the judge's own words. In most cases the safe course is an objection on the record when the material is offered.
The sentence was not an outlier
The 10.5 years was the presumptive term for the offense, and the State had alleged no aggravating factors. The prosecutor asked for less. The judge's statement that the term was "more than the State asked for" is part of what the court weighed in finding the video had an effect.
This was the second trial
The first trial ended in a new-trial order after the trial court found the State had withheld text messages from the victim's phone in violation of Brady v. Maryland and Arizona's disclosure rule. The Court of Appeals affirmed that order in a 2024 memorandum decision. Our guide to Brady claims in Pennsylvania explains how a suppression claim works here. In this appeal the court also rejected Horcasitas's challenge to the exclusion of other text messages at the second trial, which is why the conviction stands.
Does this matter in Pennsylvania?
As precedent, no. An Arizona intermediate appellate decision does not bind a Pennsylvania court, and I did not find a Pennsylvania appellate decision addressing AI-generated victim impact material. As a roadmap, yes. The principle at the core of the opinion is not peculiar to Arizona: a defendant is entitled to be sentenced on accurate, reliable information, and a judge who says on the record that a piece of evidence moved him has made the argument for the appellant. Pennsylvania sentencing judges also hear victim impact statements, and lawyers should expect to see more audio, video, and recreated material in them.
Here is what that means in practice for anyone with a Pennsylvania sentence:
- Read the sentencing transcript for what the judge said he relied on. The most important sentences are often the judge's own remarks before imposing sentence.
- Look for media. Videos, recordings, enhanced or recreated images, and statements written in someone else's voice should be identified and preserved in the record.
- Check whether an objection was made. Pennsylvania generally requires challenges to the discretionary aspects of a sentence to be raised at sentencing or in a post-sentence motion. If the objection was missed, the question may become one of counsel's effectiveness, which is a PCRA claim with its own requirements.
- Watch the clock. A PCRA petition has a one-year deadline, and a direct appeal has a thirty-day one. A new decision elsewhere does not extend either.
The broader point is the one the Arizona court made plainly. Courts are not yet equipped with settled rules for synthetic media, and a judge's emotional reaction to a polished video is not a substitute for reliability.
Questions about a sentence that may have rested on unreliable information? Request a case review or call (215) 568-0607. We check the transcript and the deadline first.
Sources
- Arizona Court of Appeals, State v. Horcasitas, No. 1 CA-CR 25-0191 (Sept. 30, 2026) (opinion of the court)
- Reuters, Arizona court says judge wrongly allowed AI-generated victim video (Sept. 30, 2026)
- ABA Journal, Judge wrongly permitted AI-generated victim video, state appeals court says
- ABC15 Arizona, Arizona appeals court throws out sentence after judge relied on AI-generated victim video
- FOX 10 Phoenix, Arizona manslaughter sentencing vacated due to use of AI 'victim impact statement'
This post is general information about a published court decision, not legal advice, and it does not create an attorney-client relationship. Every case turns on its own facts. Past results and rulings in other cases do not guarantee a particular outcome.