By WorldwideScope Technology & U.S. Desk
An Arizona appeals court has overturned a 10.5-year manslaughter sentence after a trial judge considered an artificial intelligence-generated video that portrayed a deceased victim speaking directly to the court during sentencing.
The unusual case involves Christopher Pelkey, an Army veteran who was shot and killed during a 2021 road-rage confrontation in Chandler, Arizona. His family later created an AI-generated video using his photographs, previous recordings, and a script written by his sister. The digital recreation portrayed Pelkey speaking to his family, the judge, and Gabriel Paul Horcasitas, the man convicted in his death.
However, the Arizona Court of Appeals (State v. Horcasitas, 1 CA-CR 25-0191) ruled that the AI-generated portion of the presentation was fundamentally unreliable for use in sentencing proceedings. While upholding the manslaughter conviction, the court vacated the sentence and ordered a new sentencing hearing.
The decision sets an important precedent as American courts confront a growing ethical and legal question: Can artificial intelligence recreate the words, thoughts, and emotions of someone who is deceased — and should judges be allowed to consider them?
AI Recreates a Dead Victim Speaking to the Court
The AI video was presented during Horcasitas’ sentencing hearing on May 1, 2025. Pelkey’s family sought to give the deceased victim a direct voice in the courtroom. His sister, Stacey Wales, created the digital avatar using genuine photos and voice samples of her brother, paired with a script expressing what she believed he would have said.
The resulting AI avatar addressed the courtroom directly, stating to the defendant: “In another life, we probably could have been friends. I believe in forgiveness, and a God who forgives.”
The presentation differed significantly from a traditional victim-impact statement because the words delivered by the digital avatar were neither spoken nor written by Pelkey before his death. Instead, generative software reconstructed his appearance and voice to deliver imagined thoughts created by his family.
Why Did the Arizona Court Reject the AI Video?
The Arizona Court of Appeals ruled that the admission of the AI video constituted fundamental error due to a lack of legal reliability. Writing for the panel, the court distinguished authentic photographs and audio recordings from generative AI fabrications.
Where real recordings capture actual historical events or genuine statements, the AI video presented an imagined depiction of the victim’s thoughts. The opinion noted that the video “erases the interpretive distance between the family’s belief about what the victim would have said and the victim’s own voice and opinions,” rendering the sentencing procedure fundamentally unfair.
The Judge Said He “Loved” the AI Video
A central element in the appeal was the reaction of Maricopa County Superior Court Judge Todd Lang. During the 2025 sentencing, Judge Lang remarked: “I loved that AI… I feel that that was genuine… you allowed [him] to speak from his heart.”
He then sentenced Horcasitas to 10.5 years in prison — a sentence longer than the term requested by state prosecutors. The appellate panel found that Judge Lang’s explicit reliance on the avatar’s apparent forgiveness prejudiced the proceeding enough to warrant setting the sentence aside, noting that no disclaimer could cure the error.
The Conviction Was Not Overturned
The appellate ruling strictly addresses the sentencing procedure; Horcasitas’ underlying criminal conviction remains intact.
The original case stems from a November 2021 road-rage encounter in Chandler, Arizona. Horcasitas repeatedly honked behind Pelkey’s truck at an intersection, prompting Pelkey to step out of his vehicle and approach Horcasitas. Horcasitas opened fire, killing the unarmed Pelkey. A jury previously rejected Horcasitas’ claim of self-defense and convicted him of manslaughter and endangerment.
Why the AI Video Was Different From Normal Victim Statements
Victim-impact statements are guaranteed under the Arizona Victims’ Bill of Rights, allowing surviving family members to describe their grief to the court.
However, the court found that using AI to generate a synthetic voice and face creates an undue emotional weight and an illusion of authenticity that conventional statements do not possess. Because an AI avatar appears to speak directly on behalf of the deceased, it introduces a unique risk of psychological impact on a trier of fact.
The AI Victim Was Not Simply a Recording
The Court of Appeals explicitly drew a boundary between real historical media and generative AI output. While authentic video clips embedded within the family’s presentation remained admissible, the AI-generated avatar created synthetic expressions that had never occurred in real life. The court noted that because the AI model produced new, unrecorded statements, it could not be evaluated as authentic evidence.
Family Says the Goal Was to Humanize Pelkey
Pelkey’s sister, Stacey Wales, defended her family’s decision to utilize the software, explaining that their sole intent was to humanize her brother and grant him a presence in the courtroom. While acknowledging the family’s sincere intentions, the Court of Appeals emphasized that subjective intent cannot override the strict evidentiary standard required for court proceedings.
Could This Change How Courts Use AI?
As generative media tools become accessible to the public, judicial systems face growing challenges surrounding digital evidence, deepfakes, and synthetic personas. Beyond courtroom presentations, institutions are confronting broader systemic risks, such as those highlighted in a recent Google Gemini security test and the growing risks of autonomous AI.
This decision raises foundational questions for future courtroom procedures:
- Digital Likeness: Who holds the legal right to generate a deceased person’s likeness in court?
- Evidentiary Integrity: How can courts distinguish between genuine emotional expression and algorithmic synthesis?
- Precedential Bounds: Will synthetic testimonies generated by defense teams or prosecutors face similar outright bans?
The Bigger Problem: AI Can Make Imagined Words Look Real
The Horcasitas ruling underscores a core challenge of modern generative AI: its ability to render hypothetical or imagined scenarios with convincing realism. In judicial settings, where sentencing decisions must rely strictly on reliable facts, mistaking algorithmic realism for factual authenticity poses a direct threat to due process.
What Happens Next?
Horcasitas will be returned to the Maricopa County Superior Court for a fresh sentencing hearing before a different judge, without the inclusion of the AI video. His manslaughter conviction stands.
Sources
- Official court ruling: State v. Horcasitas, No. 1 CA-CR 25-0191 (Ariz. Ct. App. Sept. 30, 2026) — Justia
- Associated Press: “Sentence tossed in Arizona case where deceased victim was depicted speaking in AI-generated video”
- CBS News: “Sentence tossed after man killed in 2021 ‘speaks’ in AI video shown in Arizona court”