Arizona Appeals Court Vacates Sentencing After Judge Relied on AI-Generated Victim Video

The Arizona Court of Appeals issued a split decision that leaves Gabriel Paul Horcasitas’s manslaughter conviction intact while vacating the 10.5‑year prison sentence imposed in May 2025. The appellate panel ordered a new sentencing hearing, citing serious due‑process concerns surrounding an AI‑generated victim‑impact video presented at the original hearing.

Horcasitas was convicted of manslaughter for the fatal shooting of Christopher Pelkey, 37, during a road‑rage confrontation at a traffic light in Chandler, Arizona, in November 2021. Pelkey was unarmed at the time of the incident.

The AI‑Generated Victim‑Impact Statement

During the May 2025 sentencing, Pelkey’s family introduced a roughly four‑minute synthetic video that depicted the deceased delivering a victim‑impact statement. The video was created using artificial intelligence to generate a likeness of Pelkey and was based on a script written by his sister.

In the video, the synthetic Pelkey spoke about forgiveness, saying that in another life he and Horcasitas might have been friends. The statement did not request a specific prison term, but the judge noted that he “loved the video” and highlighted the forgiveness expressed before imposing the maximum sentence, which exceeded the term requested by prosecutors.

Appellate Reasoning and Due‑Process Concerns

The appellate panel concluded that the video was unreliable because it presented imagined thoughts as if they came directly from the deceased victim, rather than documenting an actual statement or event. This misrepresentation, the judges wrote, undermines the integrity of sentencing proceedings.

The defence was denied a meaningful opportunity to challenge who authored the script or whether it accurately reflected Pelkey’s views. The panel identified this as a due‑process problem, noting that the court’s analysis relied on material the defence could not effectively rebut.

Prosecutorial and Victim‑Family Position

Prosecutors were aware that the family intended to address the court but claimed they did not know in advance that an AI‑generated video would be used. Victim representatives argued that Arizona law permits written, audio, or video statements without prior disclosure, a point the appellate court found insufficient to override due‑process safeguards.

• The conviction remains upheld
• The original 10.5‑year sentence is vacated
• A new sentencing hearing must be scheduled
• Future use of synthetic victim‑impact material will face heightened scrutiny

The ruling does not create a blanket prohibition on the use of artificial intelligence in courtrooms. Administrative and research applications of AI continue to expand across the judicial system, and the decision is limited to synthetic victim‑impact statements presented at sentencing.

Legal scholars note that the case may set a precedent for how courts evaluate the admissibility of AI‑generated evidence, especially when it purports to convey the voice of a deceased individual. The decision underscores the need for clear procedural rules governing disclosure, authenticity, and the right of the defence to contest such material.

For English‑speaking organizations, the immediate implication is a reminder to review evidentiary practices when employing AI tools. Companies that provide AI‑generated content for legal contexts must ensure that any synthetic statements are clearly labeled, fully disclosed to all parties, and accompanied by verifiable provenance. Failure to do so could expose courts to appeals based on due‑process violations, potentially delaying proceedings and increasing litigation costs.

Implications for Future Litigation

The Arizona decision is likely to influence other jurisdictions grappling with the admissibility of synthetic evidence. Courts may now require pre‑trial hearings on authenticity, expert testimony on the generation process, and explicit consent from all parties before such material can influence sentencing or verdicts.

In Arizona, the case will be cited in upcoming disciplinary hearings for attorneys who rely on AI‑produced exhibits, and the state bar is expected to issue advisory opinions clarifying the ethical boundaries around AI‑assisted advocacy.