A dead man forgave his killer in an AI video. An appeals court threw out the sentence
The judge said he 'loved' the clip. Arizona's Court of Appeals called it unreliable and the error fundamental.
In a Maricopa County courtroom, Christopher Pelkey told the man who shot him that he forgave him. Pelkey had been dead for more than three years. The video was made with AI, using his face and voice, from a script his sister wrote.
The sentencing judge was moved. He said he "loved" the video and that it felt "genuine," then gave Gabriel Horcasitas the maximum for manslaughter, 10.5 years.
On September 30, Division One of the Arizona Court of Appeals vacated that sentence. In State v. Horcasitas, the panel left the conviction in place and sent the case back to Maricopa County Superior Court for resentencing. Its reasoning: the video "was not reliable," and letting the judge consider it was a fundamental error that made the sentencing unfair.
Someone else's words in a dead man's mouth
Prosecutors argued the clip reflected what Pelkey's sister believed he "would have said," and that it matched his written values. The court didn't question her sincerity. Its point was simpler. The video didn't record anything Pelkey said or did. It was another person's idea of him, delivered with his face and voice, and that pairing lands with a force a written statement never would.
A law firm following the case calls it the first Arizona appellate ruling on AI-generated victim impact evidence. There will be more. Synthetic grief is cheap to make now, families will keep wanting to use it, and this ruling gives Arizona defense lawyers a clear line to point at.
More from the feed
Anthropic tells the SEC its AI may soon be speeding up its own development
Buried in Anthropic's updated IPO paperwork: the company may be close to its own threshold for AI that accelerates AI research, and it admits it has missed that kind of jump before.
This malware steals your passwords, then asks Claude how much you're worth
Cisco Talos found a Windows stealer that hands your stolen life to an AI model and asks for an appraisal. If Claude refuses, it tries a cheaper model.
Malaysia caught 30 smuggled AI servers. OpenAI's answers reportedly left by API
Customs in Kuala Lumpur found $12 million in AI hardware labeled as ordinary electronics. The same day, a report said two Chinese labs trained on GPT-6 Astra's output instead.