A new floor for AI-citation discipline

By spring 2026, the standard arc of an AI-hallucination sanction was familiar: a lawyer gets caught citing a case that does not exist, the judge issues a sternly worded order, a fine in the four or low five figures lands, and the lawyer's name shows up on a slide deck at the next CLE on generative AI ethics. Sometimes the client even still wins.

Alabama's Supreme Court decided in late April that the standard arc was not stern enough. In a published opinion dismissing the appeal of a Mobile-area trust dispute, the justices took the rare step of tossing the entire appeal because the appellate briefs filed by attorney W. Perry Hall contained, in their words, "an astounding number of invalid, inaccurate, and irrelevant citations to legal authorities." They then ordered Hall to pay $17,200 in opposing counsel's attorneys' fees and costs, referred him to the Alabama State Bar for potential discipline, and barred him from filing anything else with the court unless another attorney in good standing with the Alabama State Bar co-signs the filing.

In other words: the client lost the appeal not on the merits but because their lawyer's briefs were too unreliable to evaluate.

What the briefs actually contained

The justices reviewed Hall's appellate briefs and found a level of citation rot that went beyond the usual one or two fabricated cases. The opinion described "nonexistent or misquoted authorities" and "purported quotations from an out-of-state case which does not appear to exist." Specific examples surfaced in the coverage by Bloomberg Law, Law360, and Lagniappe Mobile:

• A quotation attributed to an Illinois trust case actually traced back to a withdrawn Indiana ruling, and the pinpoint citation pointed to a Massachusetts tax case. Three jurisdictions, three different subject areas, one quotation hopped between them like it was on a road trip.
• A citation styled as "Gilbert v. Murdock" matched only an 1802 North Carolina case about enslaved people. Whatever a 2026 Alabama trust dispute needed, it was not that.
• Another fabricated citation pointed to a Louisiana ruling about a damaged automobile. A common citation pattern in chatbot output: the case shape looks right, the citation format is clean, the underlying authority is from a completely unrelated body of law.

None of those problems require a sophisticated cite-checking workflow to catch. Opening any one of the cited cases and reading the headnote would have flagged the mismatch in under a minute. The court's frustration in the opinion is partly that the errors were not even subtle. A reader paying attention would have noticed.

The footnote that everyone is going to remember

The detail that made the Alabama Supreme Court story spread is the footnote. After opposing counsel and the court started flagging the fake authorities, Hall filed something that functioned as an apology. He acknowledged that his briefs had contained hallucinated citations, said he understood the seriousness of using AI tools without verification, and promised that "the mistake will not recur."

Inside the same footnote that delivered that apology, he cited two more cases that do not exist.

Multiple outlets, including Yellowhammer News and Bloomberg Law, called this out specifically because it is the precise behavior the apology was supposed to address. The story here is a lawyer who told the court he had learned his lesson while continuing to do the thing he was apologizing for, rather than one who tried hard and got tripped up by a tool. If a verification step had existed at any point in the apology drafting, it would have caught both new fake cites before they reached the page.

The justices did not miss the implication. Their opinion treats the recurrence inside the apology as evidence that the underlying process problem had not actually been corrected.

What the punishment actually does

Three components of the sanction order are worth flagging because they are not the usual menu.

First, the appeal was dismissed in full. In most AI-hallucination cases, the underlying case proceeds and the lawyer alone bears the consequence. Here, the court treated the briefs themselves as so defective that they could not be considered, and the client's right to appellate review evaporated as a result. That is a serious step. Appellate courts try hard to reach merits rather than punish parties for their lawyer's failings; doing the opposite here was a deliberate choice.

Second, the $17,200 is opposing counsel's fees and costs, not just a punitive fine. Opposing counsel had to spend real time identifying and documenting the fake citations and the misquoted authorities. The order makes Hall pay for that work directly. As a financial structure, this aligns the cost with the harm: making the other side check your homework is not free, and the court refused to socialize that cost.

Third, the prospective ban on filing anything in the Alabama Supreme Court without a co-signing attorney in good standing turns a one-time embarrassment into a recurring professional friction. Every future Alabama Supreme Court matter Hall handles now requires roping in another attorney willing to put their name and license on his work product. That is hard to arrange, especially after the underlying story has hit Bloomberg Law and Law360. The court also referred Hall to the Alabama State Bar for potential discipline, which keeps the matter alive past the immediate appeal.

How this fits the broader 2026 picture

The Alabama opinion arrived in the same window as a federal judge in Oregon sanctioning two lawyers $110,000 for 23 fabricated citations and eight invented quotations, which remains the largest AI-hallucination penalty on record in American legal practice. The cumulative Q1 2026 sanctions count tracked by Damien Charlotin's database crossed $145,000, with new cases logging at roughly five or six per day. Courts have moved past the "did the lawyer know" phase of the conversation. The assumption now is that every lawyer knows, every lawyer has had access to warnings since at least 2023, and every lawyer who continues to file unverified AI output is making a choice.

What Alabama added to that picture is the structural option of dismissing the underlying case. Until this opinion, the implicit deal was: the system would punish the lawyer but try to protect the client from the lawyer's mistake. Hall's apology-in-a-footnote moment let the Alabama Supreme Court explain why that deal cannot always hold. If the briefs are unreliable enough that the court cannot trust them as a basis for ruling, and if the lawyer keeps producing more fake citations even after being caught, the court at some point has to treat the briefs as a procedural nullity. That is what happened here.

The graveyard lesson

The headline is easy: another lawyer let a chatbot write a brief, did not check it, got caught, and lost. The more useful lesson is about the apology footnote.

Anyone who has spent time around AI-assisted drafting knows the seduction. The output looks clean. The citations look authoritative. The tone is confident. Even a careful lawyer can read it and think the document is mostly ready, with a few light edits to do. The verification step gets treated as a polish pass rather than the load-bearing element of the work.

Hall apparently kept making that mistake even after the court had already flagged the pattern. He sat down to write the apology, the AI helped, and the AI did the same thing it had done before. He shipped it without opening the cases. That is the moment that turned a sanctions story into a "they barred him from the court" story.

The defensive posture lawyers actually need is the inverse of how most people use these tools. The output should be assumed to contain plausible-looking fiction unless every authority has been opened, read, and confirmed to say what the brief claims. Anything less, and the next footnote you write may be the one that ends up in the Reporter.