1,600+ AI Hallucination Sanctions and Counting: The Governance Checklist That Keeps Your Firm Off the List

A researcher named Damien Charlotin keeps a public spreadsheet of lawyers who got caught citing cases that don't exist. As of July 2, 2026, it holds 1,668 documented cases worldwide. 1,163 are in the US. 653 involve a practicing lawyer — not a self-represented litigant who doesn't know better.

That number is not slowing down. And in 2026, the penalties stopped being embarrassing and started being career-ending.

If your associates are using ChatGPT or Claude for research or drafting and you don't have a written policy, this is the week to fix that.

The Tracker Just Passed 1,668 Cases — And the Penalties Are Getting Worse

For the first two years of this problem, courts mostly handed out warnings and small fines. That changed in 2026.

US courts imposed at least $145,000 in sanctions for fabricated citations in the first quarter of 2026 alone. A single Oregon matter accounted for $109,700 of that, in combined sanctions and adverse costs against one attorney. Separately, the Sixth Circuit ordered two Tennessee attorneys to pay $15,000 each — $30,000 total — in Whiting v. City of Athens, calling it the clearest federal appellate statement yet of a lawyer's duty to personally verify every citation, AI-generated or not.

Two cases this spring show where this is headed.

In April 2026, the Nebraska Supreme Court indefinitely suspended Omaha attorney Greg Lake. His appellate brief had 63 citations. 57 were defective. 20 were outright fabricated — cases that don't exist, or real cases that said the opposite of what he quoted. When justices asked him directly, in open court, whether he'd used AI, he said no. He later admitted he had. It's believed to be the first US bar discipline case to fully suspend a lawyer's license over AI-related filing errors, not just fine them.

In June 2026, a federal judge in Mississippi went further in a different direction. In Withers v. City of Aberdeen, both sides filed briefs full of hallucinated citations. Judge Sharion Aycock canceled the trial, revoked two attorneys' pro hac vice admissions, disqualified two more from the case entirely, and banned two of them from her courtroom for two years. Every lawyer of record got sanctioned. Nobody was spared for being "on the other side."

Why Small and Solo Firms Are Especially Exposed

653 of the 1,668 tracked cases involve a practicing lawyer, not a pro se filer flying blind. That should worry every managing partner who has let AI use spread through the office without a written rule for it.

Here's the actual pattern behind most of these sanctions: an associate uses AI to draft a brief, doesn't check the citations, and files it. The fabricated case gets caught by opposing counsel or a clerk. Instead of owning the mistake, the attorney denies using AI, or asks the same AI tool to "double-check" its own work — which doesn't catch anything, because a hallucinating tool has no way to know it hallucinated. By the time it reaches a judge, what started as a research shortcut looks like concealment.

Big firms have malpractice teams and standing AI committees. A 2–25 attorney firm usually has neither. That gap — heavy AI adoption, zero governance — is exactly what's showing up in the sanctions data.

What ABA Formal Opinion 512 Actually Requires

None of this is a legal gray area anymore. On July 29, 2024, the ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, and it maps directly onto the Model Rules:

• Competence (Rule 1.1): You have to understand what a generative AI tool can and can't reliably do before you use it on client work.
• Candor (Rule 3.3): Telling a court you didn't know a citation was fake isn't a defense once a lawyer's name is on the filing.
• Supervision (Rules 5.1/5.3): A verbal "be careful with AI" to your associates doesn't satisfy your supervisory duty. That's a partner-level responsibility, in writing.

Every firm in this tracker was already bound by these duties. The sanctions are what happens when a firm treats them as optional.

The 5-Point AI Governance Checklist

This doesn't require hiring a compliance officer. It requires five things, in writing, this month.

1. A one-page written AI use policy. Name the approved tools. Require disclosure of AI use on any drafted document, internally at minimum.
2. A mandatory human citation-check before filing. Every case cite gets pulled and read by a human, in Westlaw or Lexis, before it goes to court. Never ask the AI to verify its own output — it can't.
3. Client disclosure language in your engagement letter. Say, plainly, whether and how you use AI on client matters. ABA Formal Opinion 512 treats this as part of your communication duty.
4. One named AI point person. A partner, not IT, who owns the policy, fields questions, and updates it as tools change.
5. A first-48-hours plan. If a hallucinated citation is already filed, know now — not in the moment — how you correct the record with the court before opposing counsel finds it first.

Firms that skip a written policy usually aren't skipping it on purpose. Nobody sat down and decided against one. It just never got prioritized — until an associate's shortcut becomes the firm's exposure.

The Bottom Line

The lawyers in Charlotin's database weren't reckless outliers. Many were competent attorneys who trusted a tool that sounds confident even when it's wrong. The tracker crossed 1,668 cases in July. It will keep climbing.

Using AI isn't the risk. Using it without a verification step and a written policy is. That's a governance problem, and governance problems have checklists — not excuses.

Want a second set of eyes on where your firm actually stands? Request a Free Audit from 302 Digital Advisory and we'll walk through your current AI usage, drafting workflow, and exposure — no cost, no obligation.