In a September 30, 2026 opinion in State v. Horcasitas, Arizona's Court of Appeals threw out the maximum manslaughter sentence handed to Gabriel Horcasitas, finding that a sentencing judge who praised an AI recreation of the late Christopher Pelkey — scripted by Pelkey's sister and voicing forgiveness — committed "fundamental error." The conviction stands; the case goes back for resentencing.
2026-10-10 · by Moe Ameen
On September 30, 2026, the Arizona Court of Appeals vacated the sentence of Gabriel Paul Horcasitas and sent his case back for a new sentencing hearing. Horcasitas was convicted of manslaughter and endangerment for fatally shooting Christopher Pelkey, 37, during a November 2021 road-rage encounter at a Chandler stoplight near Phoenix, and in May 2025 a Maricopa County judge imposed the maximum term — reported as 10 and a half years. The appeal did not challenge the conviction, which remains in place; it challenged what the judge was allowed to consider at sentencing.
At that 2025 hearing, Pelkey's family played an AI-generated video in which a digital recreation of the dead man appeared to speak, telling the court he believed in forgiveness. The piece was built by Pelkey's sister, Stacey Wales, from a photograph run through an "old age" filter and a script she wrote — the AI "Pelkey" delivered words she imagined he would have said, not anything he actually said or recorded. It was widely described at the time as the first known instance of an AI-generated video letting a victim "speak" at a US sentencing. The sentencing judge praised it on the record, saying he "loved that AI" and calling it "genuine."
The appeals court held that admitting and relying on the video was a fundamental error that rendered the sentencing fundamentally unfair. The video, the panel wrote, "does not reflect actual events" and was "created from the imaginings of the victim's sister," and the judge's own remarks showed it had clearly influenced him. In short, a synthetic recreation of a person saying things they never said does not carry the reliability that information used to decide a prison term requires. Treat the exact sentence length (most outlets say 10.5 years) and the "first-of-its-kind" framing as press characterizations; the controlling document is the court's opinion in State v. Horcasitas.
The reversal turns on one distinction worth internalizing before you publish any AI video: a synthetic clip is a portrayal, not a record. The court's objection was not that the video was AI — it was that it put invented words in a real person's mouth and presented them as his. The honest way to use generative video is the inverse of that: an identity you actually own and consented to, saying things you actually stand behind, with a human deciding what ships. That is the exact shape [Kompozy](/) enforces. You build a face-locked [AI Influencer persona](/glossary/persona-shorts) from your own likeness — not a recreation of a real third party — and a [Persona Brief](/glossary/persona-brief) governs what that persona says, so the voice is yours by design rather than someone's guess at what you "would have said."
Crucially, nothing in Kompozy posts itself. Every generated asset — Persona Shorts and longer avatar video, carousels, quote graphics, text, blog, newsletter — is editable copy that passes a per-post review gate before it publishes across the eight social platforms plus blog and email, so adding a plain "AI-generated" disclosure to a caption or on-screen is a deliberate step, not something you discover after a batch has already fanned out. [Autopilot](/glossary/autopilot) keeps the cadence running behind that gate. The Arizona case is a warning about synthetic media presented without the provenance to back it; the workable answer for a creator is a consented identity, a reviewed queue, and clear labeling — which is how a generation-and-publishing engine built for brands is meant to run. For adjacent ground, see [California's SB 1050 synthetic-performer ad-disclosure law](/news/california-sb-1050-synthetic-performer-ad-disclosure), [where YouTube's AI disclosure and likeness rules stand in 2026](/news/youtube-ai-disclosure-likeness-rules-2026), and [Google's guidance to fact-check AI content before publishing](/news/google-fact-check-ai-content-before-publishing).
In a September 30, 2026 opinion in State v. Horcasitas, the court vacated Gabriel Horcasitas's maximum manslaughter sentence and remanded for resentencing. It left the conviction in place. The court held that the sentencing judge's reliance on an AI-generated video of the deceased victim was a fundamental error that made the sentencing fundamentally unfair.
At the 2025 sentencing, the family of victim Christopher Pelkey played a video in which an AI recreation of him appeared to speak and express forgiveness. It was made by his sister, Stacey Wales, from a photograph and a script she wrote — the AI voiced words she imagined he would have said, not anything he actually recorded. It was widely reported as the first known use of such a video to let a victim "speak" at a US sentencing.
The court's objection was not the emotion or the craft but the provenance. The video "does not reflect actual events" and was "created from the imaginings of the victim's sister," so it lacked the reliability that information used to decide a criminal sentence requires. The judge's own on-record praise showed it had influenced him.
The line is between a portrayal and a record. Synthetic video of a real person saying things they never said is the highest-risk use. The defensible pattern is an identity you own and consented to, content you stand behind, clear AI disclosure, and a human reviewing each post before it ships — which is how content engines like Kompozy are designed to operate.