A decades-old murder case in Utah is forcing courts to grapple with a thorny question: should AI-generated voice clones of deceased victims be allowed to testify in capital cases? Prosecutors in Weber County created a synthetic voice of Joyce Yost, a woman murdered in 1985, using ElevenLabs' text-to-speech technology. The defense is fighting hard to block its use, and a judge won't decide until mid-November whether the AI voice can be heard by the jury at sentencing next February.

The case involves Douglas Lovell, who abducted and murdered Yost on August 10, 1985, to prevent her from testifying against him in a rape trial. Lovell admitted to the killing during plea negotiations in 1993 and has been sentenced to death twice, though both sentences were overturned on appeal. Now prosecutors want to use the AI voice clone to present Yost's prior testimony to the jury during his upcoming sentencing hearing.

How Does ElevenLabs Create a Voice Clone?

The prosecution's approach was straightforward in theory: they fed ElevenLabs' AI software audio recordings of Yost's actual voice along with a transcript of testimony she gave at a preliminary hearing before her disappearance. The tool then generated synthesized audio of her words. But experts say this process raises serious red flags about accuracy and potential jury bias.

• Training Data Opacity: ElevenLabs does not publicly disclose what data it trained its voice synthesis models on or the specific computational adjustments it uses to refine outputs, making it impossible to independently verify the AI voice's accuracy.
• Missing Paralinguistic Information: The AI model only had access to Yost's words, not the original audio recording of how she actually spoke them, meaning inflection, cadence, and emotional tone cannot be validated.
• Psychological Bias Risk: Research shows people tend to treat AI-generated content as if it came from a real human, even when they know it's artificial, raising the risk that jurors will emotionally connect with the synthetic voice.

What Are Experts Warning About?

Lovell's defense team called two expert witnesses to challenge the AI voice's use in court. Sameer Patil, an associate professor of computer science at the University of Utah with a PhD from UC Irvine, explained how human psychology can work against the jury's better judgment.

"For example, ChatGPT. Everybody knows it's AI. There's no human involved but people are developing emotional attachments to it. People are cursing at it like they're mad. Or when they describe what it does, they describe it like it's a thing that's alive," said Patil.

Sameer Patil, Associate Professor of Computer Science at the University of Utah

Patil stressed that even when jurors are told the voice is artificial, they may unconsciously treat it as authentic testimony. He also highlighted a critical gap: because the transcript contains only Yost's words without audio, there is no way to verify whether the AI's version matches how she actually spoke during that testimony.

Kenneth Marino, an assistant professor at the University of Utah and former research scientist at Google DeepMind with a PhD in machine learning from Carnegie Mellon University, added another layer of concern about the opacity of AI voice tools.

"AI companies will generally try to use the data that's most easily publicly available. For audio, that's likely things like podcasts and YouTube videos," explained Marino, noting that ElevenLabs does not publicly disclose its training data or the weights it uses to refine its AI models. "So we essentially have no idea what data it was trained on."

Kenneth Marino, Assistant Professor at the University of Utah and Former Research Scientist at Google DeepMind

What Happens Next in the Case?

The defense team has issued a subpoena to ElevenLabs demanding details about how the company trained its voice synthesis models and what computational adjustments it made to create Yost's voice clone. As of the court proceedings, ElevenLabs had not yet responded to the subpoena.

Judge Michael DiReda indicated he expects to rule on whether the AI voice can be used by mid-November. Sentencing is currently scheduled to begin in late February 2027, though Lovell's defense team has signaled they will likely request a delay. This case marks the first time a homicide victim's prior sworn testimony, presented via an AI voice clone, has been considered for use in a potential death penalty case.

The broader implications are significant. Courts have traditionally used human proxies to read transcripts live when prior witnesses cannot testify. The use of synthetic voices raises fundamental questions about whether juries can fairly evaluate evidence when it comes from a technology whose inner workings remain hidden from public scrutiny. As AI voice tools become more sophisticated and accessible, the Yost case may set a precedent that shapes how courts handle similar requests in the future.