AI gave manslaughter victim a voice but made his killer’s sentencing unfair
An Arizona court of appeals overturned a sentence imposed for manslaughter because of content produced at sentencing showing the victim appearing as an AI-generated personality.
Defendant shot and killed a victim during a confrontation at a traffic light. A jury rejected defendant’s self-defense claim and convicted him of manslaughter. At sentencing, the trial court allowed the victim’s family to present an AI-generated video recreating the deceased victim’s appearance and voice. The presentation combined genuine recordings with synthetic statements expressing forgiveness and addressing defendant and the victim’s family. The judge praised the AI presentation, characterized it as genuine, and treated the apparent forgiveness as expressing the victim’s heart. He then imposed a 10.5-year sentence for manslaughter.
Defendant challenged the video on appeal, arguing that the sentencing judge’s reliance on it violated due process. He contended that the lifelike recreation gave imagined statements undue emotional weight and an appearance of authority and authenticity, even though the victim had never made them. Because defendant had not objected at sentencing, the appellate court reviewed the challenge for fundamental error.
The Arizona Court of Appeals affirmed the manslaughter conviction but vacated the sentence and remanded for resentencing. It held that the genuine footage embedded in the presentation was permissible, but the synthetic depiction lacked sufficient reliability because it presented the sister’s imagined statements as coming directly from the victim. Disclosure of the video’s artificial nature did not cure that attribution problem. Although the court could not determine precisely how the video affected the sentence, it found that the judge relied on it and that the resulting prejudice made sentencing fundamentally unfair.
State v. Horcasitas, No. 1 CA-CR 25-0191 (Ariz. Ct. App. Sept. 30, 2026)