AI‑Generated Video Leads to Resentencing of Arizona Road‑Rage Killer

Sat Oct 03 2026 02:14:54 GMT+0000 (Coordinated Universal Time)

An appeals court overturned a ten‑year sentence after an AI‑made depiction of the victim was shown in court, raising ethical questions about synthetic media.

In Arizona, a jury had convicted Gabriel Paul Horcasitas of shooting Christopher Pelkey. His 10‑year sentence was reversed when an appeals court ruled the judges misused an AI‑generated video of Pelkey’s family‑made impact statement. The clip, created from voice recordings and photos, portrayed a fictional version of Pelkey speaking about forgiveness. The case illustrates concerns over using synthetic media in legal proceedings.

AI‑Generated Video Leads to Resentencing of Arizona Road‑Rage Killer

An Arizona court has ordered that the 10‑year sentence of Gabriel Paul Horcasitas be overturned after a judge improperly allowed an artificial‑intelligence (AI) video of a victim’s courtroom statement to be shown during a prior sentencing hearing.

Horcasitas was convicted in 2021 by a jury of shooting 37‑year‑old Christopher Pelkey during a red‑light altercation. During last year’s sentencing, the judge reviewed an AI‑created clip that depicted Dr. Pelkey, as re‑imagined by his family, speaking about forgiveness and a hypothetical friendship. The court now finds that the clip was “crossing the line” between documenting evidence and presenting a fabricated character.

Stacey Wales, Pelkey’s sister, told the BBC that the family used voice recordings, photographs and videos to produce a synthetic likeness. Judge Todd Lang had initially praised the “innovation” of the media, noting that "I heard the forgiveness" in the clip. However, the Arizona Court of Appeals ruled it insufficiently reliable and prejudicial.

Horcasitas’ lawyer, Kristen Reller, has not given comment, and the victim’s solicitor, Jessica Gattuso, declined to respond to requests for input. The case highlights growing concerns about using AI‑rendered content in legal contexts and the difficulty of assessing its admissibility.