Juror Used ChatGPT During Deliberations — Jury Dismissed and Trial Collapsed

A juror reportedly fed trial evidence into ChatGPT during deliberations, prompting the entire jury to be dismissed and no verdict to be reached. The incident exposes urgent questions about how generative AI fits into courtroom rules designed to protect a defendant's right to a fair trial.

According to a juror's account, a panel member left the courthouse, went home, opened ChatGPT, entered summaries of testimony and evidence, and asked the system to resolve the case. The judge ultimately dismissed the entire jury, ending the trial without a verdict.

Old Rules, New Tools

Jurors have long been prohibited from conducting outside research or communicating about the case: no internet searches, no contacting witnesses, and no discussing the case outside the jury room. U.S. federal model jury instructions explicitly forbid using phones, tablets, or computers to research the trial, investigate witnesses, or communicate about deliberations.

In 2020, a federal judiciary committee strengthened those instructions and cited social platforms such as Facebook, Snapchat, and WhatsApp by name. But most instructions do not yet call out generative AI tools such as ChatGPT, Claude, or Gemini — a gap defense attorneys are likely to exploit on appeal.

Why AI Is Different

AI chatbots don't feel like search engines. They read like calm, confident interlocutors who synthesize information and offer conclusions. That demeanor makes them especially persuasive — and especially dangerous — when jurors are tempted to seek a second opinion.

“AI systems synthesize jurors' subjective descriptions of testimony into authoritative-sounding judgments, even when they are inaccurate.”

Courts have already declared mistrials over far less: jurors who Googled legal definitions or looked up a defendant's name have prompted serious remedies. Generative AI turbocharges that misconduct by interpreting evidence, weighing competing accounts, and delivering persuasive verdict-like statements — often without reliable grounding.

Problems With Outsourcing Judgment

• A juror who enters their impressions into an AI hands the system a filtered, secondhand account — not the admitted evidence.
• The model has not heard witnesses, does not apply the jurisdiction's legal standards, cannot be cross-examined, and operates with opaque reasoning.
• AI systems produce confident-sounding conclusions even when incorrect, creating a particular risk in deliberations.

In a 2024 mock-trial exercise at UNC School of Law, ChatGPT, Claude, and Grok each evaluated a criminal case and returned a unanimous "not guilty" result. The exercise wasn't an argument that AI juries are inevitable, but it showed how readily available tools can encourage outsourcing of judgment.

What Courts Should Do

Legal scholars and policymakers are calling for explicit AI language in jury instructions and stronger oversight. Naming ChatGPT, Claude, and Gemini in model jury instructions is a practical, immediate step. Broader measures — including better monitoring, clearer penalties for misconduct, and guidance on how to handle suspected AI use — will also be needed.

Enforcement will be difficult. AI assistants are now embedded in phones, browsers, and operating systems and can appear as a reflexive autocomplete. That ubiquity makes truly internet-free juries harder to guarantee and means misconduct may be underreported. Still, clarifying instructions and increasing vigilance is essential: a single juror seeking a machine's second opinion can determine a defendant's fate.

Bottom line: Courts must update jury instructions and oversight to address generative AI now, not later.