Skip to main content
AI News

PolicyNews

Arizona court’s AI-video ruling turns on reliability and the judge’s response

The panel said the judge’s reaction showed the video played a role, but the record did not establish how much it affected the 10.5-year sentence.

Policy Desk · AI News

3 min read

An Arizona courthouse exterior represents the appeals court’s resentencing ruling.

An Arizona appeals court affirmed Gabriel Horcasitas’s manslaughter conviction but vacated his sentence after finding that an AI-generated victim-impact video carried undue emotional weight at sentencing. The panel said the judge’s reaction showed the video played a role, but the record did not establish how much it affected the 10.5-year term.

The Arizona Court of Appeals, Division One, filed its decision Wednesday, Sept. 30, and ordered resentencing. Its ruling concerned a video shown after Horcasitas’s conviction, not evidence presented to a jury deciding guilt or a general question about AI-generated material in court.

The video portrayed Christopher Pelkey speaking to his family, the court and Horcasitas, including a message about forgiveness. It combined AI-generated speech, a likeness and facial expressions with authentic footage of Pelkey discussing his spiritual beliefs and fishing.

What did the AI-generated video show?

Pelkey was fatally shot during a road-rage encounter in Chandler, Arizona, in November 2021. At Horcasitas’s May 2025 sentencing hearing, Pelkey’s sister, Stacey Wales, spoke about imagining what her brother might say. The video followed her remarks.

The appeals court separated the genuine footage from the reconstructed scenes. It said the authentic clips were permissible, while the generated portion presented newly composed words and expressions as Pelkey’s. In the panel’s view, that made Wales’s interpretation appear to come directly from her brother.

Pelkey’s brother also read a statement Pelkey had written about his beliefs, values and hopes. The panel distinguished that account from the video’s reconstructed speech: the statement contained Pelkey’s own words, while the generated scenes supplied speech and expressions he had not recorded.

The court compared the video with State v. Rose, an earlier Arizona case involving photographs of a homicide victim’s young sons at his graveside. Those photographs showed an actual event, the panel said. The AI-generated scenes instead portrayed imagined thoughts and a conversation that had not occurred.

Why did the court find the presentation unreliable?

Arizona allows victims to address the court at sentencing, and judges may consider relevant information beyond evidence admissible at trial. The appeals court said due process still requires information used in sentencing to have at least minimal indicia of reliability. That standard applied even though the video was presented at sentencing, not to the jury.

The state argued the video was reliable because it reflected what Wales believed Pelkey would have said, was consistent with his written account of his beliefs and came from a close family relationship. The panel rejected those reasons as sufficient to treat the generated words and expressions as Pelkey’s own.

Horcasitas argued that the video gave the victim’s imagined statements undue emotional weight and an apparent authority and authenticity that could affect the judge. The appeals court concluded that, in this case, the AI reconstruction crossed into undue prejudice and made the sentencing proceeding fundamentally unfair.

Free newsletter

Get the morning briefing

Start each day with the AI stories that matter and why — a short, free email from our newsroom.

Free. One email a day, one-click unsubscribe. See our privacy policy.

The decision addressed this video as victim-impact material at sentencing. The panel noted that no Arizona case had considered an AI-generated depiction of a victim offered for that purpose. It did not set a general rule barring AI-created material from courts or decide whether it could be admitted as evidence at trial.

Wales, who created the video, told 404 Media that she wrote the avatar’s words herself to convey her brother’s imagined perspective and would present the same script herself at resentencing if needed. Wales told the outlet: “Chris's sentiment does not change. The delivery method will”.

What did the judge’s response establish?

The trial judge, Todd Lang, praised the video and described the forgiveness it portrayed as genuine. The appeals panel treated his remarks as evidence that the presentation affected him, while also saying the record did not establish precisely how much weight he gave it when setting the sentence.

Lang imposed the presumptive 10.5-year manslaughter term. The prosecutor had requested a slightly mitigated nine-year term, and the defense had asked for a mitigated seven-year term. The opinion said the state alleged no aggravating factors.

The difference between the prosecutor’s request and the sentence does not show that the video alone caused the gap. The court did not calculate the video’s influence or assign it a number of years; it found the proceeding unfair and sent the case back for resentencing.

Horcasitas’s conviction remains in place, and only the sentence was vacated. The opinion does not set a date for resentencing, and the available reporting does not establish what victim-impact presentation will be used. The court’s stated concern was specific: generated words and expressions attributed an imagined message to Pelkey, and the trial judge’s response showed that the presentation played a role.

Related coverage

More from Policy

More Policy