Arizona Court Vacates Sentence Over AI Video of Dead Victim

Arizona Court of Appeals Vacates a Sentence Over an AI Video of the Deceased Victim

On September 30, 2026, the Arizona Court of Appeals held that a sentencing judge committed fundamental error by hearing and relying on an AI recreation of a homicide victim speaking in court. The manslaughter conviction stands. The 10.5-year sentence is vacated, and the case goes back for resentencing.

Information last verified on October 2, 2026. This is a developing story; we update it as the record changes.

Jurisdiction scope: This article addresses Arizona criminal sentencing procedure and the use of AI-generated victim impact material in Arizona state courts. It does not address how AI-generated exhibits are authenticated at trial, federal sentencing practice, or the law of any other state.

What Happened

Gabriel Paul Horcasitas shot and killed a man during a road rage confrontation in Maricopa County. According to the opinion, Horcasitas repeatedly honked his horn while stopped behind a truck at a red light; the victim got out, raised his arms, and advanced toward Horcasitas's car while yelling; within seconds, Horcasitas shot him (paragraph 2).

Horcasitas told police the same day that the victim had charged at him, repeatedly threatened to kill him, and grabbed the frame of his open driver's-side window. Several eyewitnesses contradicted that account, reporting that the victim never advanced past the rear of his own truck and that nobody saw him touch or reach into the car (paragraph 3).

The case has a long procedural history. A first jury acquitted Horcasitas of first-degree felony murder, second-degree murder, and drive-by shooting, convicting him of the lesser-included offense of manslaughter and one count of endangerment. The trial court then granted a new trial after finding the State had violated both Brady v. Maryland and Arizona Rule of Criminal Procedure 15.1(b)(8) by withholding text messages from the victim's phone, and the Court of Appeals affirmed that order in an April 2024 memorandum decision (paragraphs 6 through 8).

Before the second trial, Horcasitas pleaded guilty to one count of endangerment. A second jury then convicted him of manslaughter as a dangerous offense (paragraph 12), rejecting the justification defense he had raised under Arizona's rules on deadly force and self-defense.

The sentencing hearing

Judge Todd F. Lang of the Maricopa County Superior Court imposed the presumptive term of 10.5 years on the manslaughter count and a concurrent presumptive term of 2.25 years on the endangerment count (paragraph 13).

At the hearing, numerous friends and family members of the victim spoke. Several asked for the maximum sentence the law allowed, which, because the State had alleged no aggravating factors, was the presumptive 10.5-year term for a class 2 dangerous felony under A.R.S. section 13-704(A). The victim's brother read a lengthy statement the victim had written himself (paragraph 35).

The victim's sister then spoke. She told the court she constantly imagined what her brother would have said if he could have taken her place, because he was not there to say it himself (paragraph 35). At that point, the AI video played.

The video opened with the AI figure explaining that it was a version of the victim recreated through AI using his picture and his voice profile. Real footage of the victim was embedded inside it. The AI figure then said the video was a true representation of who he was, "not how the courtroom portrayed" him, thanked the court and the people in attendance, and addressed Horcasitas directly, saying in part, "I believe in forgiveness and in God who forgives" (paragraph 36).

The State asked for a slightly mitigated nine-year term, citing Horcasitas's lack of criminal history. Defense counsel asked for seven years (paragraphs 38 and 40).

The sentencing judge told the family, "I loved that AI. Thank you for that," said he "heard the forgiveness," and added that he felt "that was genuine." He described the family as having let the victim "speak from his heart as you saw it," and noted, "I didn't hear him asking for the maximum sentence." He then found mitigating circumstances, including a law-abiding life and remorse, but concluded that the family's loss "reduces the effectiveness or power" of that mitigation, and imposed the presumptive 10.5 years, observing that this was "more than the State asked for" (paragraph 41).

The ruling

Presiding Judge D. Steven Williams wrote for a unanimous panel that also included Judge Anni Hill Foster and Vice Chief Judge David D. Weinzweig. The court affirmed the manslaughter conviction, rejecting Horcasitas's separate argument about excluded text messages, and vacated the manslaughter sentence "because the AI video was not reliable" (paragraph 1).

A note on naming: the opinion identifies the deceased only as "the victim," redacts even his first name where the sentencing judge used it, and the appellate docket lists the crime victims by initials. We follow the court's convention throughout this article and do not name him.

What the Law Actually Says

Arizona victims have a constitutional right to be heard, and the family holds it when the victim is dead

Article 2, section 2.1(A)(4) of the Arizona Constitution gives crime victims the right "[t]o be heard at any proceeding involving a post-arrest release decision, a negotiated plea, and sentencing." Subsection (C) defines "victim" to include, when the person is killed or incapacitated, the person's spouse, parent, child, or other lawful representative.

A.R.S. section 13-4426 implements that right. Subsection (A) lets the victim "present evidence, information and opinions that concern the criminal offense, the defendant, the sentence or the need for restitution" at any aggravation, mitigation, presentencing, or sentencing proceeding. Subsection (B) gives the victim the right to be present and to address the court at any disposition proceeding.

The court spelled out how that right operates in practice. Under A.R.S. section 13-4426.01, a victim exercising the right to be heard at sentencing does not have to disclose the statement to the State, the defendant, or the court in advance. The victim is not sworn, and the statement is not subject to cross-examination (paragraph 42). Nothing in the standard trial machinery for testing a statement applies.

The court also restated the limit on that right. Victims' rights are to "be liberally construed," but the right to speak "is not unlimited," and victim statements "cannot infringe on the defendant's constitutional rights, including the right to due process" (paragraph 43, quoting Lynn v. Reinstein).

Sentencing evidence in Arizona is loose, but it is not unlimited

The rules of evidence do not govern a sentencing hearing. Arizona Rule of Criminal Procedure 26.7(b), as the court quoted it, lets any party introduce "any reliable, relevant evidence, including hearsay, to show aggravating or mitigating circumstances." A sentencing judge has "wide discretion in the sources and types of evidence" used (paragraphs 44 and 45).

The due process floor is reliability. Quoting State v. McGill, the court held that only evidence bearing a "minimal indicia of reliability" may be considered at sentencing, because "defendants have a due process right to a fair sentencing procedure which includes the right to be sentenced on the basis of accurate information" (paragraph 45, quoting State v. Grier).

Because Horcasitas did not object to the video when it was played, the panel reviewed only for fundamental error, the stricter standard (paragraph 34). Victim impact evidence is permitted at sentencing "unless it is so unduly prejudicial that it renders" the proceeding "fundamentally unfair," the test drawn from State v. Rose.

Why this particular video failed

The court said plainly that "[n]o Arizona case has addressed the admissibility of an AI-generated depiction of a victim offered as victim impact evidence" (paragraph 48). It worked by analogy to Rose, where the Arizona Supreme Court considered photographs of a murder victim's young sons at his gravesite, called them "troubling" and "uncomfortably close" to the line, and still found no error, reasoning that even "without the photographs" it was evident "that the murder caused the two boys to suffer a devastating loss" (paragraph 49). The Horcasitas court drew the contrast itself: those photographs, it said, "captured a particularly poignant moment for the young surviving victims," unlike the AI video (paragraph 50).

The AI video, the panel held, crossed that line for two reasons.

First, it did not depict anything that happened. Setting aside the embedded authentic clips, the video "presents a depiction of the victim and his thoughts created from the imaginings of the victim's sister" (paragraph 50).

Second, it presented those imaginings as the victim's own speech. The opinion is worth quoting directly: 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," and the AI figure's own claim to be "a true representation of who I was" made that worse. "No disclaimer could cure that error" (paragraph 51).

The court then pointed to the record on effect. The judge said he "loved" the video, felt it "was genuine," and relied on the AI figure's "obvious forgiveness." While the record "does not reflect precisely how the AI video factored into the sentencing calculus, there is no question it played a role" (paragraph 52).

That combination, fundamental error plus demonstrated prejudice on this record, produced the remedy: the manslaughter sentence is vacated and the case is remanded for resentencing (paragraphs 53 and 54).

What the court did not do

It did not disturb the conviction. It did not vacate the concurrent 2.25-year endangerment sentence. It did not hold that AI-generated material is categorically inadmissible, in Arizona courts or anywhere else. And it expressly approved the authentic video footage of the victim that was embedded in the presentation.

The separate evidentiary issue

The opinion also resolves a second question that has nothing to do with AI, and it is the part that kept the conviction intact. Horcasitas argued the trial court wrongly excluded text messages the victim's brother-in-law had sent the victim, which described him as perceiving everything as aggressive and prone to snapping while driving.

The court held the messages were inadmissible as a matter of form. Under Arizona Rule of Evidence 405(a), a victim's character trait can be proved only by testimony about reputation or testimony in the form of an opinion, not by extrinsic evidence such as text messages, and the messages were too general to count as "other acts" under Rule 404(b) (paragraphs 22 and 23). A separate text in which the victim expressed despair three days before the shooting was admissible hearsay under Rule 803(3) as a statement of then-existing state of mind, but the panel held the trial court acted within its discretion in excluding it under Rule 403, because ample and uncontroverted eyewitness evidence had already established that the victim was the initial aggressor (paragraphs 27 through 32).

Status: published, and not yet final

The decision is captioned "Opinion of the Court" and the docket entry for it reads "OPINION," so this is an opinion of the court rather than a memorandum decision. The docket records the disposition as "Affirmed in Part, Vacated, Remanded in Part," filed September 30, 2026, and lists two open due dates: a motion for reconsideration is due October 15, 2026 and a petition for review to the Arizona Supreme Court is due October 30, 2026. No case-closed date appears on the docket we read. Readers can pull the docket themselves through the Arizona courts' public access portal; see our guide to how to search Arizona court records for what that portal covers.

Analysis: Why This Matters

The following is analysis from the Recording Law Editorial Team.

The court's holding is narrow. It did not rule on whether AI-generated video can be evidence. It ruled that this AI artifact, a synthetic recreation of a dead person delivering first-person speech he never spoke, lacked the minimal indicia of reliability that due process requires of what a judge considers at sentencing. The gate the opinion names is reliability: "We vacate the manslaughter sentence and remand for resentencing because the AI video was not reliable" (paragraph 1). The court said no Arizona case had addressed the admissibility of an AI-generated depiction of a victim offered as victim impact evidence, and it decided only the case in front of it (paragraph 48).

In our reading, paragraph 51 is the load-bearing one. The family said openly what the video was, and the video itself announced that it was an AI recreation, so concealment is not the ground the court gives. The ground it gives is attribution: the video "presents the statement (including non-verbal communications made through facial expressions and the like) as coming directly from the victim" and "erases the interpretive distance between the family's belief about what the victim would have said and the victim's own voice and opinions" (paragraph 51). A sister saying she believes her brother would have forgiven the defendant is a statement of her own belief, and a court can weigh it as such. The same words delivered by a synthesized rendering of the victim's face and voice arrive instead as the victim's own position. On what labeling can fix, the opinion says only this: "No disclaimer could cure that error" (paragraph 51).

That distinction matters to how the Arizona Victims' Bill of Rights works in practice. Family members speaking for a victim who was killed fall within the constitutional definition of "victim" in article 2, section 2.1(C), and the court did not question their right to be heard; it restated that the right to speak "is not unlimited" and that victim statements cannot infringe the defendant's constitutional rights, including the right to due process (paragraph 43). What this opinion addressed was form. It treated the family's beliefs about the deceased, offered as the family's beliefs, differently from the same content staged as the deceased's own first-person statement (paragraphs 50 and 51).

The statutory posture of this channel is part of the picture. By statute, a victim's sentencing statement need not be disclosed in advance to the State, the defendant or the court, is not made under oath, and is not subject to cross-examination (paragraph 42). The opinion does not say whether the defense saw this video beforehand, or why no objection was made.

The appellate posture also cuts against reading this case too broadly. Horcasitas did not object, so the court reviewed only for fundamental error (paragraph 34). It found both fundamental error and, "on this record" (paragraph 53), prejudice, pointing to the judge's own statements that he "loved" the video and felt it "was genuine" (paragraph 52). A different sentencing record, with a judge who said nothing about an AI presentation, could come out differently on prejudice even under the same reliability holding.

How This Affects You

If you are following this specific case, the practical state of play is simple. The conviction is intact, the manslaughter sentence is not, and a Maricopa County judge will conduct a new sentencing hearing. We will not speculate about what term will be imposed, and neither should any coverage you read. Arizona's sentencing range for a first-time class 2 dangerous felony runs from 7 years to 21 years under A.R.S. section 13-704(A), with 10.5 years as the presumptive term; what happens within that range on remand is for the sentencing court.

If you are a crime victim or a victim's family member in Arizona, the right to be heard at sentencing comes from Ariz. Const. art. 2, section 2.1(A)(4) and A.R.S. section 13-4426, and this opinion restated that the right to speak "is not unlimited" and cannot infringe the defendant's constitutional rights, including the right to due process (paragraph 43). What the court held is that this AI video, which presented the sister's imagined account as the victim's own words, should not have been allowed at this sentencing (paragraphs 50 through 53).

If you came to this story because you are trying to understand the wider rules on synthetic likenesses, that is a different body of law from sentencing procedure. The state statutes aimed at cloned voices and digital replicas, which we track in our guide to AI voice cloning and digital replica statutes, govern commercial and harassing uses rather than courtroom admissibility. Arizona's own AI legislation is collected in our overview of what Arizona has actually enacted on artificial intelligence.

This is general legal information, not legal advice. It covers Arizona state criminal sentencing procedure and reflects sources verified on October 2, 2026. Laws change and this story is developing; consult a lawyer licensed in your jurisdiction about your specific situation.

Related articles

• Arizona's rules on deadly force and self-defense
• What Arizona has actually enacted on artificial intelligence
• AI voice cloning and digital replica statutes, state by state
• How to search Arizona court records

---

Last updated: 2026-10-02. This is a developing story; details verified as of 2026-10-02.

Frequently Asked Questions

Did the Arizona Court of Appeals overturn the conviction?

No. The court affirmed the manslaughter conviction. It vacated only the sentence imposed for manslaughter and remanded for resentencing (State v. Horcasitas, No. 1 CA-CR 25-0191, paragraph 54). The concurrent 2.25-year endangerment sentence was not disturbed.

Does this mean the defendant is being released?

No. A vacated sentence is not an acquittal and not a release. The Arizona conviction stands and the case returns to the Maricopa County Superior Court for a new sentencing hearing.

Can an AI-generated video be used as evidence in court?

This opinion does not answer that question in general. It addressed one narrow situation, an AI recreation of a deceased victim speaking, offered as victim impact evidence at an Arizona sentencing hearing, where the rules of evidence do not apply and the due process standard is a 'minimal indicia of reliability' (paragraph 45). Authentication of AI-generated exhibits at trial is a separate question this opinion did not reach.

Why was the AI video a problem if everyone knew it was AI?

The court said disclosure did not fix it. The video presented the sister's beliefs about what her brother would have said as the victim's own speech and expressions, which 'erases the interpretive distance' between the two, and 'no disclaimer could cure that error' (paragraph 51).

Was any of the video allowed?

Yes. The court held that the authentic footage of the victim embedded inside the presentation 'was permissible' and that only the AI rendering constituted fundamental error (paragraph 53).

Do Arizona crime victims still have the right to speak at sentencing?

Yes. Ariz. Const. art. 2, section 2.1(A)(4) gives victims the right to be heard at sentencing, subsection (C) defines 'victim' to include, when the person is killed or incapacitated, the person's spouse, parent, child or other lawful representative, and A.R.S. section 13-4426 lets a victim present evidence, information, and opinions about the offense, the defendant, and the sentence. The court restated that the right to speak 'is not unlimited' and that victim statements cannot infringe the defendant's constitutional rights, including the right to due process (paragraph 43).

What standard of review did the court apply?

Fundamental error, because Horcasitas did not object to the video at sentencing (paragraph 34). The panel found both fundamental error and prejudice sufficient to render the sentencing procedure fundamentally unfair on this record (paragraph 53).

Is this a published opinion, and is it final?

It is an opinion of Division One of the Arizona Court of Appeals rather than a memorandum decision, per the docket entry reading 'OPINION' and the 'Opinion of the Court' caption. It is not final. The court's docket lists two open due dates, with a motion for reconsideration due October 15, 2026 and a petition for review to the Arizona Supreme Court due October 30, 2026, and shows no case-closed date.

Why is sentencing evidence treated so differently from trial evidence?

Arizona Rule of Criminal Procedure 26.7(b) lets any party at a presentence hearing introduce 'any reliable, relevant evidence, including hearsay,' and a sentencing judge has wide discretion over sources of information (paragraphs 44 and 45). The limit is due process: only evidence with a minimal indicia of reliability may be considered, because a defendant has a right to be sentenced on accurate information.

Where can I read the opinion myself?

The opinion is published on the Arizona Court of Appeals Division One website as State v. Horcasitas, No. 1 CA-CR 25-0191, filed September 30, 2026. The official appellate docket for the case is available through the Arizona Judicial Branch appellate case lookup. Both are linked in the sources for this article.