When AI hallucinates in court, the lawyer pays the price

Both sides, same mistake

In a civil rights case in Mississippi, Judge Louis Guirola Jr sanctioned the lawyers for both the plaintiff and the defendant, and cancelled the trial ^[1]. Both sides had filed briefs citing cases that do not exist. The lawyers had used AI to draft parts of their submissions, and the AI had hallucinated the citations.

It is a bit like handing in a school essay with a bibliography you never checked. The teacher spots three books that were never written, and gives you an F. Except here, it is a federal court, and the sanction goes on your permanent record.

The Mississippi case is not an outlier. In Pennsylvania, the Middle District suspended attorney Anselmo Mancini for a year after he submitted a brief with six fake cases from ChatGPT, then claimed they were real, then blamed a paralegal ^[2]. The court said the problem was not using AI; it was failing to check the output and then lying about it.

What the rules say

The American Bar Association's Model Rule 1.1 has always required lawyers to provide competent representation ^[3]. Comment 8 to that rule now says a lawyer must keep up with changes in the law and its practice, "including the benefits and risks associated with relevant technology".

That does not mean lawyers cannot use AI. It means they have to understand what it can and cannot do, and they remain responsible for everything filed in their name.

In these cases, the AI worked exactly as designed: it predicted plausible-sounding text, including case names, citations and quotes. It did not check whether those cases exist. That is the lawyer's job.

The evidence question

When a brief lands on a judge's desk full of citations that turn out to be fiction, the question is simple: did the lawyer check them, or did they trust the AI?

In the Mississippi case, both sides appear to have trusted the output without verifying it. The judge found that out when he tried to read the cases ^[1].

In the Pennsylvania case, the lawyer first said the cases were real, then said a paralegal had done the research, then admitted he had used ChatGPT and not checked the results ^[2]. The suspension followed not just the mistake, but the attempt to hide it.

The lesson is that a printout from an AI is not proof that a human reviewed it. The question, then, is how a lawyer shows they did the checking.

What this means for you

If you use AI to help draft legal work, or any professional work where accuracy is not optional, here are three questions worth asking:

• Can you show which parts were AI-assisted? If a question comes up later, you will want to be able to point to where the AI was involved and where you checked it.
• Do you have a record of your review? A note, a tracked change or an email saying "checked all citations" is better than nothing, but it is still just your word. A sealed record of each version, with a note of what you checked and when, is harder to dispute.
• What happens if the AI gets it wrong? The rules put the responsibility on you, not the tool. The question is whether you can prove you did your part.

A witness to the checking

This is where a digital witness fits well. The thing to seal is not the AI's output, but the record of your review.

Say you use AI to draft a section of a brief. You check each citation, correct two that are wrong and add a note: "All citations verified against Westlaw, 15 December 2024." You seal that note, together with a fingerprint of the final version.

Months later, if someone questions a citation, you can show that the version you filed is exactly the one you reviewed on that date, and that your note of the checking hasn't changed. The AI's role is clear, and so is yours.

Sealing happens through WitniumChain: your system takes a fingerprint of the note and the document, sends it via the REST API, the SDK or the MCP server, and gets a witness ID back. To check it, you calculate the fingerprint again and compare. The content stays with you; only the fingerprint is sealed.

What it shows is that the record of your work existed at that time and hasn't changed. What it doesn't show is that every citation was correct, or that your review was thorough. But it does show that the review happened, and when.

Anyone you share the record with, including a court, can check it without having to take your word for it. That is the point.

How it would work

1. Draft with AI assistance, review the output and note what you checked.
2. Seal the final version and the review note when you file the brief.
3. If questioned, show the sealed record: unchanged since the day you did the work.

What to watch

The AI Act's transparency rules apply from 2 August 2026 for most systems ^[4]. Article 50.4 says that AI-generated text published to inform the public on matters of public interest must be disclosed, unless it has been reviewed by a person with editorial responsibility.

That is aimed at news and public information, not legal briefs, but the principle is the same: when AI writes something, someone has to check it, and it helps to be able to show that they did.

The first 2,000 witnesses a month are free. The documentation is at witniumchain.com/docs.

Sources

1. Lawyer sanctioning and trial cancellation in Curry v. Collins et al., reported by JD Supra, December 2024. United States District Court for the Southern District of Mississippi. https://www.jdsupra.com/legalnews/using-ai-as-a-writing-assistant-not-a-9962930/

2. In re Anselmo Mancini, United States District Court for the Middle District of Pennsylvania, No. 1:23-mc-00189, Memorandum and Order filed 30 October 2024. Attorney suspended for one year for submitting AI-generated hallucinated citations and attempting to conceal it. Reported by Legal Dive, November 2024. https://www.legaldive.com/news/chatgpt-ai-hallucinations-律师-suspension-sanctions/732111/

3. Model Rules of Professional Conduct, Rule 1.1: Competence, American Bar Association. Comment 8 added in 2012 addresses technology competence. https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_1_1_competence/

4. Regulation (EU) 2024/1689 on artificial intelligence (AI Act), Article 50.4 on transparency obligations for deployers of certain AI systems, European Commission AI Act Service Desk. https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50