What Error Did Arizona’s
Court of Appeals Say
No Disclaimer Could Cure?

The AI video began by describing itself as a version of the victim “recreated through AI,” and the sentencing judge said, “I loved that AI.” One panel of the Arizona Court of Appeals vacated the manslaughter sentence on September 30, 2026, and wrote: “No disclaimer could cure that error.”

During the sentencing hearing in Maricopa County Superior Court, a woman told the court, of her brother, “I imagine it constantly because I know he would say everything I would absolutely never be able to say myself.” The Arizona Court of Appeals records that remark in an opinion filed September 30, 2026, and says that at that point an AI video was played. In a report published September 30, 2026, FOX 10 Phoenix identifies the sister as Stacey Wales. The opinion does not name her. The court affirmed the manslaughter conviction and vacated the sentence.

The case is State of Arizona v. Gabriel Paul Horcasitas, No. 1 CA-CR 25-0191. The opinion calls the brother only “the victim”; ABC News, on October 7, 2026, names him Christopher Pelkey. The opinion states that Horcasitas shot and killed the victim during a road rage confrontation, and that a jury rejected his claim of self-defense and convicted him of manslaughter. It records that the superior court sentenced him to a presumptive term of 10.5 years’ imprisonment on the manslaughter count.

What the court decided

One three-judge panel of Division One affirmed the manslaughter conviction and vacated the sentence, writing that it vacated “because the AI video was not reliable.” Reviewing only for fundamental error, because Horcasitas did not object to the AI video at sentencing, the panel held that allowing the presentation of the AI video was fundamental error and that, on this record, the judge’s consideration of it rendered the sentencing procedure fundamentally unfair. The embedded “real” footage of the victim, the court wrote, was permissible.

The court called the question a “novel issue”: the opinion states that no Arizona case had addressed the admissibility of an AI-generated depiction of a victim offered as victim impact evidence.

How the video opened

In paragraph 36, the opinion describes the opening. The video began with the AI victim explaining that it was a “version” of himself “recreated through AI” that used his picture and voice profile. Auto-generated captions of the family’s upload (audio not heard) open the same way, but we cannot say that upload is the version played in court, so every word we quote from the video comes from the opinion.

What the judge said

The opinion’s caption names the superior court judge in the case under appeal as Todd F. Lang. The opinion quotes him telling the family, “I loved that AI.” In the same passage he said, “I feel that that was genuine,” and told them they had allowed the victim “to speak from his heart as you saw it.” The opinion does not explain “as you saw it.” The appeals court wrote that the video “clearly impacted the sentencing judge,” and also: “While the record does not reflect precisely how the AI video factored into the sentencing calculus, there is no question it played a role.”

Through October 9, 2026, we read in full the ten sources listed below other than the docket. None states whether, or by how much, the video changed the term imposed. We read article text only. We did not read the sentencing transcript or the briefs.

What the court wrote about the form

The defense argued the video carried “undue emotional weight” and conveyed “an authority and authenticity” with a “psychological impact” on the sentencing judge. The State argued it was reliable on three grounds, the first being that it reflects what the victim’s sister “believed” he “would have said,” had he spoken. The opinion records both as arguments.

Paragraph 51 addresses the explanations. The court wrote that “despite the explanations concerning the genesis of the AI video, it presents the statement (including non-verbal communications made through facial expressions and the like) as coming directly from the victim.” In its words, the video “erases the interpretive distance between the family’s belief about what the victim would have said and the victim’s own voice and opinions.” It quoted the AI victim saying, “this here is a true representation of who I was, you know, not how the courtroom portrayed me.” Then it wrote: “No disclaimer could cure that error.” On our reading, the explanations the court mentions include the video’s own opening, though the opinion does not list them and the briefs, which we did not read, may contain others.

The court also described where the depiction came from. The video, it wrote, presents “a depiction of the victim and his thoughts created from the imaginings of the victim’s sister.” The court had italicized “imagine” in her remarks, fifteen paragraphs earlier in the opinion. Wales told 404 Media, in an article dated October 6, 2026: “The words I penned as Chris’s imagined opinion still have not changed and will not change.” Her account reaches us through that interview, not the court record, and her words are consistent with the court’s phrase. Neither is a finding about who generated any audio or image. 404 Media also quotes her saying, of a resentencing: “Chris’s sentiment does not change. The delivery method will.”

In its statement of Arizona law, the court wrote that victims at sentencing do not appear as witnesses. It also wrote that they take no oath before making statements to the court and that their statements are not subject to cross-examination.

The Agentics read

Agentics reads this record through Digital Derangement Syndrome™ (DDS™): AI-mediated systems failing to recognize legitimate authority. On our mapping, not the court’s, the closest fit is Trust Transfer Failure (our term for expertise that exists but has never been translated into machine-readable trust), in the inverse direction: a rendering wore a person’s voice. In the court’s words, the video “erases the interpretive distance” between the family’s belief and the victim’s own voice. A professional whom a labelled system renders in the first person, in a reply sent in their name, a biography in their voice or a post under their byline, meets the same grammar: the label says what the rendering is, while its sentences still arrive as “I.” That is an argument, not evidence. The fit is partial, it says nothing about what anyone in the courtroom understood, and we make no claim that any professional has been harmed.

Where the argument stops

As of the court’s September 30, 2026 docket, a motion for reconsideration is due October 15, 2026 and a petition for review to the Arizona Supreme Court is due October 30, 2026: deadlines, not filings.

The opinion, read in full on October 9, 2026, does not state what would make an AI-generated victim statement permissible, and neither do we. The auto-generated captions of the family’s upload (audio not heard), read from 0:00 to 3:37 of a 231-second video (frames, on-screen text and the closing seconds not covered), contain no spoken line naming who wrote the words. In a report updated October 5, 2026, CBS News/AP said the defense argued to the appeals court that the video does not disclose who wrote the words used by the AI version of Pelkey; the opinion’s account of the defense’s arguments does not record that one. We offer neither as a finding about the video.

The mapping to Trust Transfer Failure is an editorial judgment, not the court’s or any source’s finding.