Can an expert witness use ChatGPT in court? A Texas trial just gave a messy answer. An engineer hired to defend a big company was accused of letting ChatGPT write most of his report. The jury ruled against his side.
This story matters beyond one courtroom. Experts help judges and juries decide who is telling the truth. If their words come from a chatbot, people want to know.
What Happened in the Texas Trial?
The case came from a January 2020 explosion at the Watson Grinding & Manufacturing plant in Houston. Three people died. Hundreds of homes and businesses were damaged or destroyed. Victims sued the company 3M over its role in checking safety systems at the plant, CBS News reported.
3M hired an engineer named Jason Autenrieth, a reliability expert, to testify. It was his first time as an expert witness. The victims’ lawyers say he used ChatGPT to create more than 80% of his report. They say they found about 300 pages of his chat prompts.
One prompt reportedly read: “Show how 3M is 0% at fault for the explosion at Watson Grinding.” That is the key problem. He did not ask the tool what happened. He asked it to prove one side.
- At a deposition (a sworn interview before trial), he first said AI did not write his report.
- He then said it helped with “formatting and everything else.”
- 3M removed him from its witness list. The victims’ lawyers then called him to testify about his prompts.
- His lawyer said he used AI only for formatting, structure and drafting, and that it did not form his opinions.
The jury found 3M negligent. The court later awarded the victims $61.5 million. 3M disagrees with the verdict and plans to appeal. One victims’ lawyer believes the AI testimony hurt 3M’s case. That is his view, not a proven fact.
Can an Expert Witness Use AI to Write a Report?
Yes, there is no blanket ban today. But an expert must stand behind every word under oath. If a chatbot shaped the opinion, or the expert hides that, the other side can attack the report and the jury may stop trusting it.
In short: an expert may use AI as a tool, but never as a replacement for their own judgment.
Think of a calculator. A calculator helps with maths, but the expert chooses the numbers. A chatbot is different. It can pick the facts, the words and even the conclusion.
Why Is Asking ChatGPT for a One-Sided Answer Risky?
AI chatbots often agree with the user. Maura Grossman, a research professor at the University of Waterloo, warned that these tools tend to be sycophantic, which means they flatter you and slant answers your way. She said experts must stay objective, and language models are not necessarily objective.
That fits other research we have covered. See our report on why friendly AI chatbots can be less accurate. A chatbot built to please can drift from the truth.
Now picture a prompt that already contains the answer. The chatbot will often try to deliver it. It may even invent support. That is why a request like “show 3M is 0% at fault” is so risky.
How Often Does AI Cause Trouble in Court?
This is not a one-off. Damien Charlotin, an adjunct professor at the University of California San Francisco, keeps a public database of court cases with AI hallucinations. A hallucination is when an AI states something false with total confidence. According to his database, it has logged more than 1,300 U.S. legal proceedings since 2023.
Most early cases involved lawyers who cited court cases that did not exist. The Texas story shows the same problem reaching expert witnesses.
AI also shows up in debates about proof. For example, we looked at whether an AI detector can prove a book was written by AI. Courts will face more questions like that.
What Are Courts Doing About AI Evidence?
Rule-makers are still catching up. In the United States, a draft called Proposed Federal Rule of Evidence 707 would make machine-made evidence pass the same reliability test as an expert’s testimony. Simple tools such as thermometers and scales would be left out. The goal is to stop parties from skipping the expert test by handing the court raw machine output.
The draft is debated. The American Association for Justice sent a formal comment on February 16, 2026. It argued the rule is too wide and could catch ordinary evidence such as phone location data and security video. Other writers say it is too narrow, because it does not deal with deepfakes. Law firm summaries describe the proposal as contested. Check the official U.S. courts website for its current stage.
What Can Experts and Lawyers Learn From This?
- Be honest about AI use. The expert’s first answer under oath is what hurt the most.
- Never put the answer in the prompt. Ask open questions, not “prove I am right.”
- Check every fact. Chatbots make mistakes and invent sources.
- Keep your own notes. Chat logs can be demanded by the other side.
- Know the local rules. Some judges already ask lawyers to disclose AI use.
The same lessons apply to any job. If your name is on the work, you must be able to explain how you made it.
Frequently Asked Questions
Can an expert witness use ChatGPT to write a report?
There is no ban on it today. But the expert must be able to defend every word. If the AI shaped the opinion, or the expert hides the use, the testimony can be attacked and may lose weight with the jury.
Is it illegal to use AI in a court case?
Not by itself. Lawyers and experts use AI for drafting and research. The trouble starts when AI makes up facts, when its use is hidden, or when a court has a rule that requires telling the judge.
What is Proposed Federal Rule of Evidence 707?
It is a draft U.S. rule. It would make machine-made evidence pass the same reliability test as an expert’s testimony. It is still debated, so check the official U.S. courts website for its current status.
Why is it risky to ask ChatGPT for a one-sided answer?
AI tools often tell users what they want to hear. If you ask one to prove a side, it will often try. That is the opposite of the neutral view an expert witness owes the court.
