Knowledge Center

AI Hallucination Cases in Court: What the 2026 Data Shows

Reported AI hallucination cases in court leveled off in 2026. What the data shows, what it cannot show, and the new verification rules.

Reported AI hallucination cases in court stopped growing in 2026. Damien Charlotin’s database counts 460 decisions in Q1 2026 and 372 in Q3, as of November 16, 2026, and recent quarters fill in late. Law Tech AI reads that as a plateau with a modest dip, and the data cannot show that new verification rules caused it.

On October 23, 2026, Reuters ran the headline “AI ‘hallucinations’ in court filings show signs of slowdown, but judges still sound alarms.” A common reading followed: the new verification rules for lawyers are working. The data supports the slowdown and says nothing yet about the cause.

Picture a managing partner at a 20-attorney firm who reads that headline and asks whether the firm can drop its second-reader check on briefs. The answer starts with what the numbers measure.

Are AI hallucination cases in court going down?

The count stopped climbing in early 2026 and has dipped modestly since. Damien Charlotin’s AI Hallucination Cases database logged 411 decisions in Q4 2025, 460 in Q1 2026, 434 in Q2 and 372 in Q3, as of its November 16, 2026 update.

Charlotin saw the plateau early. In May 2026 he told Scientific American: “For the past two or three months, we have reached a plateau of around 350, 400 decisions a quarter.”

The dip is small. Compared at the same point after each quarter closed, Q3 2026 showed 366 decisions about four weeks after September 30 (archived October 31). Q2 showed 396 about four weeks after June 30 (archived August 3). That is about 8 percent fewer.

Why do the most recent months look so low?

The database records court decisions, which arrive months after the filing and then have to be found and entered. The newest quarter always looks smaller than it will end up.

On June 30, 2026, the database showed 337 decisions for Q2. By November 16 it showed 434 for the same quarter, almost 30 percent more.

Q4 2026 shows 96 decisions through mid-November. That number will rise, and any claim built on it is premature.

Are lawyers making fewer AI citation errors, or only pro se litigants?

Both groups declined, and the mix barely moved. Lawyers account for about 39 to 40 percent of decisions in each of the first three quarters of 2026, and self-represented litigants for 57 to 60 percent, per the database’s own party breakdown.

By Law Tech AI’s count from the database’s public CSV, decisions involving lawyers went from 180 in Q1 2026 to 167 in Q2 and 153 in Q3. Decisions involving self-represented litigants went from 273 to 255 to 210.

Attorney verification rules reach lawyers first, so if they were driving the change, the lawyer share would fall. Through Q3 2026 it held at about 40 percent, and the pro se count fell faster.

Did new verification rules cause the slowdown?

The data cannot answer that, and the timing points away from most new rules. The plateau was visible by May 2026, before the rules people are citing took effect.

  • New York’s statewide rule, 22 NYCRR Part 161, took effect June 1, 2026. It requires no AI disclosure and offers judges a model rule.
  • California’s SB 574 starts January 1, 2027, and Pennsylvania’s proposal is still open for comment.

Other explanations fit the same numbers, and none has been tested: tools that check citations, judges handling repeat errors without a written decision, filers taking more care, or fewer decisions reaching the database. The count also has no denominator, because nobody tracks the total number of AI-assisted filings.

Which courts require lawyers to verify AI citations now?

The duty itself is old. Federal Rule of Civil Procedure 11 and its state counterparts already treat a signature as a certification that legal contentions are warranted by existing law. What is new is written language that names AI.

  • New York, June 1, 2026. Where a judge adopts the Part 161 model rule, signing a paper “certifies that such a review has been conducted and that the paper contains no such fabricated or fictitious content.”
  • Ninth Circuit, December 1, 2025. An advisory committee note to Circuit Rule 32-1, quoted in Lnu v. Blanche, says: “Regardless of how the filing is prepared, the signature is an attestation that the signer has reviewed the filing and is responsible for the accuracy of its contents.”
  • California, January 1, 2027. SB 574 bars court papers from carrying a citation the responsible attorney “has not personally verified.” Our SB 574 post covers the statute.
  • Pennsylvania, proposed. 56 Pa.B. 6888 (October 24, 2026) would have every filer, self-represented litigants included, certify review of any AI-drafted language. Comments are due November 24, 2026, to AIAC@pacourts.us.

Individual judges add standing orders on top. Check the judge’s rules at intake.

What should a firm do with this data?

Keep the check. A Q3 count of 372 still works out to about four decisions a day, and every one involves a filing that someone signed.

The rules are moving toward written certification. Jennifer Case, J.D., a California-licensed attorney and MCLE provider who founded Law Tech AI, recommends that firms using ChatGPT for lawyers or any other AI tool write down three things now:

  1. The reviewer who reads every cited authority against a primary source before filing
  2. What that reviewer records in the matter file
  3. What the firm does the day an error is found

The first two follow ABA Formal Opinion 512 and our post on preventing AI hallucination sanctions. The third belongs in your firm’s AI policy before SB 574 takes effect on January 1, and we will cover it in our next post.

Put the review step in writing

If your firm’s AI policy stops at the approved-tool list, the verification and incident steps come next. We can work through them with you. Book a discovery call at law-tech.ai.

FAQ

How many AI hallucination cases have courts reported?

As of November 16, 2026, Damien Charlotin’s AI Hallucination Cases database lists 2,305 court and tribunal decisions worldwide, 1,565 of them in the United States. Among decisions where AI use was found, 400 carried a monetary sanction and 172 a disciplinary referral. The database counts decisions that address hallucinated content, so filings that no court wrote about are absent from the total.

Are AI hallucinations in court filings decreasing?

The count leveled off in 2026 and dipped modestly. Charlotin’s database shows 460 decisions in Q1 2026 and 372 in Q3, as of November 16, 2026. Compared at the same point after each quarter closed, Q3 ran about 8 percent below Q2. Recent quarters keep growing as late decisions are added, so the newest numbers understate the final count.

Do lawyers or pro se litigants file more AI hallucinations?

Self-represented litigants account for more of the reported decisions. In each of the first three quarters of 2026, they made up about 57 to 60 percent of decisions in Charlotin’s database, and lawyers about 39 to 40 percent, as of November 16, 2026. Both groups declined between Q1 and Q3 2026, and the lawyer share stayed near 40 percent.

Why do recent months show so few AI hallucination cases?

Decisions arrive months after the filing that caused them, and each one has to be found and entered. On June 30, 2026, Charlotin’s database showed 337 decisions for Q2 2026. By November 16, 2026, it showed 434 for the same quarter. Counts for the latest month or quarter will rise, so they cannot show a trend yet.

Which courts require lawyers to certify AI citations?

New York’s Part 161 model rule, effective June 1, 2026, treats a signature as certifying that AI-assisted papers were reviewed, where a judge adopts it. California’s SB 574 requires personally verified citations in court filings starting January 1, 2027. Pennsylvania proposed a certification for all filers in October 2026. Many individual judges also have standing orders, so check each judge’s rules.


Jennifer Case, J.D., is the founder of Law Tech AI and a LawNext columnist. She is a California-licensed attorney (Bar #223008) and a California MCLE provider (#20734). She works with small to mid-size law firms on practical and safe AI adoption. Visit law-tech.ai.

This post describes the data and the rules as of November 16, 2026. It is general information. For advice on a specific matter, talk to counsel.

Jennifer Case

Jennifer is a California-licensed attorney (Bar #223008) and the founder of Law Tech AI. A former litigator turned entrepreneur, she ran and coached businesses for two decades, including 18 years as a CEO, before building Law Tech AI to help attorneys adopt artificial intelligence safely, ethically and effectively.

Read next
  1. Nvidia’s CEO Resigns, Anthropic’s Redacted Review, Claude Meetings, and Pennsylvania’s AI Rule

    Anthropic says it may soon cross its own threshold for AI that speeds up AI research, and it redacted most of the key evidence in the outside review it invited.…

  2. Democrats Take Congress, Claude’s Legal Tools, the NYT Sues Brave, and Billing for AI Work

    Democrats took both chambers of Congress in a blue wave, and the fight over whether a federal AI law should override state laws now goes to a new majority. Voters…

  3. The AI Questions Law Firm Candidates Ask in Interviews

    Firms recruiting AI-forward associates and paralegals need ready answers about AI policy and training. Seven questions to prepare for.