AI victim impact statements in sentencing: what the Arizona appeals court held, and what a dated file can show

This page is for a general audience, including families and lawyers who have seen news of these cases. It is not legal advice and it is not a comment on anyone named except as the official opinion, or a named news report, already describes them. A broader TimeBinder guide to deepfake evidence rules is Deepfake evidence and the rules: US federal practice, England and Wales, and the EU AI Act.

What the Arizona court held

The official opinion is State v. Horcasitas, No. 1 CA-CR 25-0191, filed on 30 September 2026 by the Arizona Court of Appeals, Division One. Presiding Judge D. Steven Williams delivered the opinion, joined by Judge Anni Hill Foster and Vice Chief Judge David D. Weinzweig. The appeal was from the Superior Court in Maricopa County, the Honourable Todd F. Lang, Judge.

As the opening paragraph of the opinion states, Horcasitas shot and killed the victim during a road-rage confrontation. A jury rejected his claim of self-defence and convicted him of manslaughter. At sentencing, the superior court heard and relied on an artificial-intelligence video rendering of the victim speaking to his family, the court, and Horcasitas, which the court characterised as “genuine” and as speaking from his heart. The Court of Appeals affirmed the conviction. It vacated the manslaughter sentence and remanded for resentencing because the AI video was not reliable.

The opinion describes the video in some detail. It began with an AI depiction of the victim explaining that the video was a “version” of himself “recreated through AI” that used his picture and voice profile. A real video of the victim was embedded in that presentation. When that ended, the AI depiction returned and said the video was “a true representation of who [he] was … not how the courtroom portrayed [him].” It then addressed Horcasitas, including a statement about forgiveness, and ended with remarks to friends and family, alongside what appeared to be another embedded real video. Because Horcasitas had not objected to the video at sentencing, the appellate court reviewed only for fundamental error.

The court distinguished photographs of surviving relatives at a gravesite, which the Arizona Supreme Court had considered in State v. Rose. Unlike those photographs, the AI video (excluding the embedded real footage of the victim) did not reflect actual events. It presented “a depiction of the victim and his thoughts created from the imaginings of the victim’s sister.” Despite explanations of how the video was made, it presented the statement, including non-verbal expressions, as coming directly from the victim, and “erase[d] the interpretive distance between the family’s belief about what the victim would have said and the victim’s own voice and opinions.” The opinion says no disclaimer could cure that error.

The sentencing judge, the opinion records, said he “loved” the video and felt it was genuine, pointing to the AI depiction’s “obvious forgiveness.” The appellate court said the record did not show precisely how the video factored into the sentence, but that there was no question it played a role. It held that the embedded real footage of the victim was permissible, but that allowing the AI video was fundamental error, and that consideration of it so prejudiced Horcasitas as to render the sentencing procedure fundamentally unfair. The manslaughter sentence — a presumptive term of 10.5 years, which the opinion notes was more than the State had asked for — was vacated and the case remanded for resentencing. The conviction stands.

A different reported pattern in England

A different fact pattern was reported in England shortly beforehand. According to a BBC News report published on 23 September 2026, the Court of Appeal rejected a sentence appeal by Jerome Gibson after it emerged that a police officer had used Microsoft Copilot to create a statement. The BBC’s own report describes that document both as a witness statement and, in the judges’ remarks, as a personal statement. It reported that the officer, who had been removed from front-line duties, instructed the tool to create “the most impactful personal statement,” to “make a judge or reader weep,” and to help secure the highest sentence, and that Mr Justice Lavender, sitting with Lord Justice Edis and Mr Justice Linden, said the facts in the statement were not AI-generated hallucinations but things the person referred to as F had stated in her interview. As reported, the court said it did not appear that the use of AI, “deplorable though it is,” caused the sentence to be any longer than it otherwise would have been, and that the total sentence was not manifestly excessive. Derbyshire Police, the BBC said, confirmed a criminal investigation into an allegation of perverting the course of justice in a number of cases, with no arrests made at the time of the report.

That account comes from the BBC’s report of the hearing. We have not seen the official judgment, so the description here is attributed to that report; check the official text before relying on it in a case. The reported contrast with the Arizona opinion is still useful: one case concerns an AI video presented as the victim speaking; the other, as reported, concerns an AI-drafted written statement whose underlying facts the court treated as the victim’s own, while still condemning the officer’s method.

What this does and does not mean

Neither decision is a general ban, in every court, on every use of software in connection with a victim statement. The Arizona opinion is a holding about reliability and due process on the facts of that sentencing, including a video that claimed to be a true representation of the deceased victim. The English appeal, as reported, condemned an officer’s use of a chatbot to write a statement but did not disturb the sentence on the record before that court. Other courts may be asked to draw their own lines. For the wider rules on authenticating recordings and dealing with deepfakes as evidence, see the guide to deepfake evidence rules in the US, England and Wales, and the EU.

For families, the Arizona court did not say that real recordings or writings of the person, or live statements by surviving relatives, were out of bounds. It said the embedded real footage of the victim was permissible, and the opinion recounts that family members also spoke in their own words and that the victim’s brother read a statement the victim had actually written. The line the court drew was at an AI depiction offered as the victim’s own voice and thoughts.

This page does not tell anyone how to persuade a judge, and it does not comment on the family in Horcasitas beyond what the opinion itself records.

What a dated file can show

A TimeBind is a dated, checkable record that an exact file existed no later than a recorded time and still matches. The fingerprint is made in your browser and the file stays on your device. You get a Proof of Time Certificate. Your first Pooled TimeBind is free, with no credit card required. After that, TimeBinds start from US$5, and an ASAP TimeBind is US$15.

If a person, while alive, wrote a letter, recorded a message, or saved a video of themselves, a TimeBind of that exact file supports the account that those bytes existed by that date and still match. That is a different object from an AI video generated later from a picture and a voice profile. Dating the AI file only dates the AI file. It does not turn it into a statement by the person depicted, and it does not cure the reliability problem the Arizona court described.

TimeBinder does not authenticate a courtroom exhibit, does not advise on admissibility, and does not detect whether a video was generated. Lawyers who are preserving digital evidence have a separate guide for lawyers and a legal timestamping hub for law firms; this page is not that guide.

Sources checked

• State v. Horcasitas, No. 1 CA-CR 25-0191 (Ariz. Ct. App. Div. 1, filed 30 September 2026), official opinion (court website).
• BBC News, “Police officer used AI to create statement – court,” 23 September 2026 (the official England and Wales judgment was not retrieved from Find Case Law during this check; facts in that section are attributed to the BBC report).
• TimeBinder, “Deepfake evidence and the rules: US federal practice, England and Wales, and the EU AI Act”.

Start a dated record of a file the person actually made

Your first Pooled TimeBind is free, with no credit card required. After that, TimeBinds start from US$5, and an ASAP TimeBind is US$15. Click here to start, or see plans and pricing.