Skip to content

Vol. I · No. 18October 2026

A register of developments in law and legal tech, sourced and in full.

Court order · Ariz.

Arizona appeals court vacates manslaughter sentence over reliance on AI victim video impact statement

The Arizona Court of Appeals affirmed Gabriel Horcasitas's manslaughter conviction but vacated his 10.5-year sentence and remanded for resentencing, holding that the trial judge's consideration of an AI-generated video of the deceased victim, Christopher Pelkey, rendered the sentencing fundamentally unfair. The three-judge panel found the video, scripted by Pelkey's sister to depict him addressing the court and forgiving the defendant, lacked sufficient reliability.

The Arizona Court of Appeals has thrown out a manslaughter sentence because the judge heard and relied on an AI-generated video of the dead victim forgiving the defendant. In a published opinion filed September 30, 2026, Division One affirmed Gabriel Paul Horcasitas’s conviction but vacated his 10.5-year sentence and sent the case back for resentencing, holding that the video “was not reliable” and that considering it made the sentencing “fundamentally unfair” (opinion, ¶¶ 1, 53). Presiding Judge D. Steven Williams wrote for the panel, joined by Judge Anni Hill Foster and Vice Chief Judge David D. Weinzweig. The case is State v. Horcasitas, No. 1 CA-CR 25-0191, on appeal from Maricopa County Superior Court, where Judge Todd F. Lang presided.

How the case got here

According to the opinion, Horcasitas shot and killed the victim during a road rage confrontation at a red light. The opinion does not name the victim. At the first trial, the jury acquitted Horcasitas of first-degree felony murder, second-degree murder and drive-by shooting, and convicted him of manslaughter and one count of endangerment. The trial court then granted a new trial because the State had not disclosed text messages from the victim’s phone, which the court found violated Brady v. Maryland and Arizona Rule of Criminal Procedure 15.1. The Court of Appeals affirmed that order in an April 4, 2024 memorandum decision (¶¶ 5 to 8).

Before the second trial, Horcasitas pleaded guilty to endangerment. A second jury convicted him of manslaughter as a dangerous offense. The court imposed the presumptive term of 10.5 years for manslaughter and a concurrent presumptive term of 2.25 years for endangerment (¶¶ 12 to 13).

The video

At sentencing, family and friends described the victim, his brother read a statement the victim had written about his beliefs, and his sister asked what he “would have said if he could have taken my place today” (¶ 35). An AI video then played. It opened with the AI figure describing itself as a “version” of the victim “recreated through AI” using his picture and voice profile, included real footage of him, and had the figure call itself “a true representation of who I was, you know, not how the courtroom portrayed me.” It told Horcasitas, “I believe in forgiveness and in God who forgives” (¶ 36).

The family’s lawyer asked for the maximum. Because the State alleged no aggravating factors, that was the presumptive 10.5 years. The prosecutor asked for 9 years and the defense for 7 (¶¶ 35, 37 to 40). Judge Lang told the courtroom, “I loved that AI,” said he felt it “was genuine,” and said the family had “allowed [the victim] to speak from his heart as you saw it.” He found mitigating factors, said the family’s loss reduced their weight, and imposed 10.5 years, adding, “That’s more than the State asked for” (¶ 41).

Why the panel said the video crossed the line

Horcasitas did not object at sentencing, so the panel reviewed only for fundamental error (¶ 34). It started from the Arizona Victims’ Bill of Rights, which gives victims the right to be heard at sentencing, without an oath, cross-examination or advance disclosure (¶ 42, citing Ariz. Const. art. 2, § 2.1(A)(4) and A.R.S. §§ 13-4426, 13-4426.01). That right “is not unlimited,” the court said, and due process still requires that sentencing evidence carry a “minimal indicia of reliability” and that a defendant be sentenced on accurate information (¶¶ 43, 45).

The State argued the video was reliable because it reflected what the victim’s sister believed he would say, the two were close, and it matched his own written statement (¶ 47). The panel noted that no Arizona case had addressed an AI-generated depiction offered as victim impact evidence, and it worked from the Arizona Supreme Court’s 2013 decision in State v. Rose, which found gravesite photographs of a victim’s sons “uncomfortably close” to the line but not over it (¶¶ 48 to 49).

The AI video, the panel held, went over. Unlike photographs, it “does not reflect actual events” but “presents a depiction of the victim and his thoughts created from the imaginings of the victim’s sister” (¶ 50). 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,” and “No disclaimer could cure that error” (¶ 51). The judge’s own words showed it mattered: while the record does not show exactly how the video factored in, “there is no question it played a role” (¶ 52). The real footage embedded in the video was permissible; the AI portions were not (¶ 53).

The conviction stands

The panel rejected Horcasitas’s challenge to the exclusion of text messages at the second trial (¶¶ 14 to 32). Texts from the victim’s brother-in-law describing the victim as quick to “snap” were general character opinions, not specific acts, and Arizona Evidence Rule 405(a) allows such opinions only through live testimony. The earlier ruling that the texts had to be disclosed did not decide whether they were admissible. A text from the victim himself three days before the shooting fit the hearsay exception for state of mind, but the trial court acted within its discretion in excluding it under Rule 403, because eyewitnesses uniformly described the victim as the initial aggressor.

Who it applies to

The ruling binds the parties and, as a published opinion of the Court of Appeals, can be cited to Arizona courts as precedent, unlike the 2024 memorandum decision in the same case. It addresses an AI video presented as victim impact evidence at a sentencing hearing, reviewed for fundamental error because no one objected. It vacates only the manslaughter sentence; the endangerment sentence was not challenged and is not disturbed.

What happens next

The court’s docket lists a motion for reconsideration due by October 15, 2026, and a petition for review to the Arizona Supreme Court due by October 30, 2026 (docket, as of September 30, 2026). Absent further review, the case returns to the Maricopa County Superior Court for resentencing on the manslaughter count.

What is not yet known

The sources reviewed do not say whether the State or the victims’ counsel will seek reconsideration or review, or when resentencing will be held. The opinion does not decide whether an AI depiction could ever be presented at sentencing if disclosed in advance and challenged, though it says no disclaimer could have cured this one. It does not address AI-generated material offered as evidence at trial. Whether Arizona’s courts will adopt a rule or guidance on AI-generated victim statements is not addressed in the sources reviewed.

Sources

The Arizona Court of Appeals, Division One opinion of September 30, 2026, and the court’s docket for No. 1 CA-CR 25-0191; reporting by Reuters, FOX 10 Phoenix and AZFamily.

Filed under
Court orderAriz.The story on LexCircuit

Sources6 articles across 4 outlets