Part of the AI Regulation News hub.
Arizona's Court of Appeals vacated a manslaughter sentence and remanded for resentencing because the judge heard and relied on an AI-generated video of the victim, while affirming the conviction itself
The video told the court it had been made with AI. The panel said that did not help, because the AI portion still presented the family's imagined words as coming from the victim himself, while the real footage embedded in it was permissible.
Bottom line: This is a filed opinion of an intermediate state appeals court, Arizona's Court of Appeals, Division One, dated 30 September 2026, and it binds the parties. The manslaughter conviction is affirmed. The manslaughter sentence is vacated and the case goes back to the superior court for resentencing. The ruling concerns the sentencing phase in Arizona only.
Who this affects: Arizona criminal defense lawyers, prosecutors and victim-advocacy counsel preparing victim impact presentations, and sentencing judges who screen what victims and families play in court.
Issue date: Opinion filed 09-30-2026 in No. 1 CA-CR 25-0191, on appeal from Maricopa County Superior Court No. CR2021-142720-001. The opinion sets no deadline for the resentencing.
What changed: The panel held that "allowing the presentation of the AI video constituted fundamental error" and found that, on this record, the judge's consideration of it rendered the sentencing procedure fundamentally unfair. The opinion says no Arizona case had previously addressed an AI-generated depiction of a victim offered as victim impact evidence.
Analysis: On our reading, the useful line for practitioners is the split inside one exhibit. Embedded real footage of the victim was permissible; the AI portion was not, and the court said its built-in explanation that it was an AI recreation did not save it. The opinion does not announce a ban on AI victim presentations.
Primary sources: State v. Horcasitas, No. 1 CA-CR 25-0191, Opinion (PDF, Arizona Court of Appeals, Division One)
- Instrument (EN)
- State of Arizona v. Gabriel Paul Horcasitas, Opinion of the Court, No. 1 CA-CR 25-0191
- Authority
- Arizona Court of Appeals, Division One. Presiding Judge D. Steven Williams wrote the opinion; Judge Anni Hill Foster and Vice Chief Judge David D. Weinzweig joined
- Jurisdiction
- Arizona (state court), appeal from the Superior Court in Maricopa County, the Honorable Todd F. Lang
- Status
- Decided. Affirmed in part; vacated and remanded in part. Conviction affirmed; manslaughter sentence vacated; remanded for resentencing
- Bindingness
- Binding on the parties. A published intermediate appellate opinion; we did not check whether any party has sought further review
- Issue date / next deadline
- Filed 09-30-2026. No resentencing date appears in the opinion
- Standard of review
- Fundamental error, because the defendant did not object to the AI video at sentencing
- Primary source
- https://coa1.azcourts.gov/Portals/1/OpinionFiles/Div1/2026/State%20v.%20Horcasitas%20-%201%20CA-CR%2025-0191%20-%20Opinion.pdf
What did the court decide?
The panel affirmed the manslaughter conviction and vacated the manslaughter sentence. The opinion's caption reads "AFFIRMED IN PART; VACATED AND REMANDED IN PART", and its conclusion states: "we affirm the manslaughter conviction but vacate the sentence for manslaughter, and remand for resentencing." This is a filed opinion of Arizona's intermediate appeals court, and the case now returns to the superior court for a new sentencing.
The panel rejected the challenge to the conviction on evidentiary grounds. Horcasitas argued that the trial court wrongly kept out text messages from the victim's phone. The panel held that texts from the victim's brother-in-law were inadmissible character evidence under Evidence Rules 404 and 405, and that a text from the victim himself, although admissible as a state-of-mind statement under Evidence Rule 803(3), was properly excluded under Evidence Rule 403 on this record.
The panel vacated the manslaughter sentence. The superior court had imposed the presumptive term of 10.5 years for manslaughter, a dangerous offense, and a concurrent presumptive 2.25 years on an endangerment count to which Horcasitas had pleaded guilty. The disposition names only the manslaughter sentence.
What was in the video, and what did the judge say about it?
According to the opinion, the victim's sister spoke at the sentencing hearing, and then a video played that depicted the "victim" speaking to the camera. It opened with the AI figure explaining it was a "version" of himself "recreated through AI that use[d] [his] picture and [his] voice profile." Real footage of the victim talking about his faith was embedded in it. The AI figure then said the video was "a true representation of who [he] was . . . not how the courtroom portrayed [him]", and told Horcasitas, "I believe in forgiveness and in God who forgives."
The judge's later remarks are quoted at length. They include "I loved that AI" and "I feel that that was genuine", and the judge added that he "didn't hear him asking for the maximum sentence." He found mitigating circumstances, said the family's loss "reduces the effectiveness or power of the mitigating fact that Mr. Horcasitas has led a lawful life", and imposed the presumptive term, which he noted was more than the State asked for. The prosecutor had sought 9 years; defense counsel sought 7.
Horcasitas did not object at the hearing. That is why the panel reviewed only for fundamental error.
Why did the AI portion fail when the real footage passed?
The panel started from Arizona law that gives victims a constitutional right to be heard at sentencing, and from the rule that sentencing evidence needs a "minimal indicia of reliability" to satisfy due process. It then compared the video with the gravesite photographs the Arizona Supreme Court reviewed in State v. Rose. Those photographs captured a real moment. The AI video, the panel wrote, "(excluding the embedded "real" videos of the victim) does not reflect actual events" and instead presents a depiction "created from the imaginings of the victim's sister."
The disclosure at the start of the video did not change the result. In the panel's 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", and "No disclaimer could cure that error."
On prejudice, the court conceded some uncertainty and still found it enough: "While the record does not reflect precisely how the AI video factored into the sentencing calculus, there is no question it played a role." It concluded that, "on this record," the judge's consideration of the video "so prejudiced Horcasitas as to render the sentencing procedure fundamentally unfair."
What does the opinion leave open?
The opinion leaves much open. The holding is tied to this video and this record, at a sentencing hearing, in Arizona. The opinion does not say that every AI-assisted victim presentation is barred, and it expressly treats the embedded real footage as permissible. On our reading, a family's own spoken or written account of what the victim would have wanted is not what the panel faulted; the opinion's objection is to presenting that belief as the victim's own speech.
The opinion contains no concurrence and no dissent in the text we read. It also does not set a procedure for screening AI material before a hearing, although it repeats the Arizona Supreme Court's encouragement in Rose that judges "screen and, if necessary, limit" unduly prejudicial victim impact presentations.
For defense counsel, the posture matters: this result came on fundamental-error review after no objection was made. For prosecutors and victim-advocacy counsel, the State's arguments that the video reflected what the sister believed, rested on a close relationship and matched the victim's written statement were all before the panel, and it still found the AI portion unreliable.
What we did not verify
What we opened: the full 16-page opinion PDF from the Arizona Court of Appeals, Division One website, as text extracted from that PDF, including the caption, counsel listing, all numbered paragraphs 1 to 54 and the clerk's filing block. Every quotation on this page comes from that file, with typographic quotes folded to straight quotes and words rejoined where a line break split them.
What we did not open: the superior court record, the video itself, the sentencing transcript beyond the passages quoted in the opinion, the parties' briefs, the filing by Arizona Voice for Crime Victims, and the 2024 memorandum decision in the earlier appeal. We did not check the docket for any petition for review to the Arizona Supreme Court or for a resentencing date.
What we refuse to claim: that the conviction was overturned, that Arizona courts have banned AI-generated victim impact evidence, that the ruling reaches trials or any court outside Arizona, or that the panel decided a broader constitutional question beyond the fundamental-error and due-process reliability analysis it set out. We do not say what sentence will follow on remand, and we make no claim about the endangerment sentence beyond what the disposition says.
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
On our reading of this Arizona opinion, an AI rendering that speaks as a deceased victim at sentencing carries real risk to the sentence, even with an on-screen explanation that it is AI. Real footage of the victim stood. Counsel preparing or opposing a victim impact presentation would do well to separate recorded material from generated speech, and to raise the issue at the hearing rather than leave it to fundamental-error review.
Source File
Open the opinion in No. 1 CA-CR 25-0191 and confirm the filing date of 09-30-2026 and the caption disposition on page 1, the summary at paragraph 1, the analysis of the AI video at paragraphs 48 to 53, and the conclusion at paragraph 54 affirming the conviction and vacating the manslaughter sentence.
In sum, while the embedded "real" video footage of the victim was permissible, allowing the presentation of the AI video constituted fundamental error. ยท State v. Horcasitas, Arizona Court of Appeals, Division One, No. 1 CA-CR 25-0191, para. 53, filed 30 September 2026
FAQ
Did the Arizona Court of Appeals overturn the conviction?
No. The panel affirmed the manslaughter conviction, rejecting the argument about excluded text messages. It vacated the manslaughter sentence and remanded for resentencing.
Does the opinion ban AI victim impact videos in Arizona?
It does not announce a categorical ban. It held that presenting this AI video was fundamental error and, on this record, rendered the sentencing procedure fundamentally unfair. It said the embedded real footage of the victim was permissible.
Why didn't the video's AI disclosure solve the problem?
The panel said the video still presented the statement as coming directly from the victim and erased the distance between the family's belief about what he would have said and his own voice. It wrote that "No disclaimer could cure that error."
Why was the error reviewed as fundamental error?
Because Horcasitas did not object to the AI video at sentencing. The panel reviewed only for fundamental error and found prejudice on this record, noting that the judge said he "loved" the video and felt it "was genuine".
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.