Expensive Counsel, Cheap Verification

Butler Snow was not some solo practitioner trying to bluff through a filing with a chatbot and a deadline problem. It was one of the firms Alabama had paid tens of millions of dollars to defend the state's prison system in a long-running stream of litigation. That made the July 23, 2025 sanctions order from U.S. District Judge Anna Manasco harder to dismiss as an isolated embarrassment.

The underlying case involved an inmate who alleged prison officials had failed to keep him safe after he was stabbed multiple times at William E. Donaldson Correctional Facility. That is serious litigation on its own. Add the larger context of Alabama prison oversight, violence allegations, and federal scrutiny, and this was exactly the kind of case where you would want the lawyers to check whether the authorities in their brief existed.

They did not.

Judge Manasco found that Butler Snow lawyers had submitted filings that relied on fake authorities generated through ChatGPT-assisted research. The problem was not subtle. The citations were, in the court's words, "completely made up." The order did not describe a close interpretive dispute or a formatting mistake. It described fabricated legal support attached to actual court filings.

What the Court Found

According to the reporting and the sanctions order, an attorney used artificial intelligence to research supporting law for two filings and then failed to verify the results before inserting them into the papers. The false citations reached federal court in a live case. When the problem surfaced, Butler Snow lawyers apologized and acknowledged the role of AI. By then the damage was already done.

Courts do not care whether a fake citation comes from carelessness, panic, or an LLM that talks like a second-year associate with a head injury. The filing lawyer is still the filing lawyer. The signature on the brief still means the same thing it always meant: I checked this.

Judge Manasco's order was notable because she treated the failure as more than a garden-variety oops. She called the conduct "recklessness in the extreme." She also made clear that fabricated authority is not a paperwork problem. It is misconduct that distorts the adversarial process and wastes the court's time. Opposing counsel has to chase ghosts. Judges have to verify fiction. Everyone's calendar gets worse because someone wanted a faster first draft.

The Sanctions Went Further Than a Fine

The sanction package mattered as much as the misconduct. Manasco publicly reprimanded William R. Lunsford, Matthew B. Reeves, and William J. Cranford. She removed all three from participation in the case. She also ordered them to circulate the sanctions order to clients, opposing lawyers, and judges in their other pending matters. Then she referred the incident to the Alabama State Bar for possible discipline.

That disclosure requirement is the part that makes this story unusually radioactive. A routine sanction hurts. A sanction you must proactively hand to other judges and other clients travels.

Once a court orders that kind of disclosure, the reputational damage stops being confined to one docket. Every future opposing counsel now has an obvious reason to scrutinize the firm's citations. Every judge receiving the order has an obvious reason to wonder how many other filings deserve a closer look. Every client gets a fresh reminder that the expensive outside counsel defending a politically sensitive state institution fed make-believe cases into federal court.

The profession often talks as if the risk from AI hallucinations is primarily the sanction itself. That is too narrow. The more lasting cost is the collapse of presumed reliability. Lawyers live on the assumption that a signed brief is at least attached to reality. Once that assumption cracks, everything gets slower and more suspicious.

Why This Was Worse Than Another ChatGPT Shame Post

By mid-2025, no lawyer in the United States could credibly claim ignorance about fake AI citations. The 2023 Avianca sanctions had already become the legal profession's most durable cautionary tale. Bar associations, judges, legal media, and every anxious litigation partner in America had spent two years repeating the same boring instruction: if a machine gives you a citation, look it up before you file it.

Butler Snow still managed to blow through that warning in one of the least forgiving contexts available: prison litigation, under federal scrutiny, representing the state, with real people and real claims on the other side.

There is also a structural point here. Large firms like to present AI misuse as an individual lapse. That framing is convenient. It suggests the institution is sound and one person simply had a bad idea. But once a federal judge is removing multiple lawyers from a matter and ordering broad downstream disclosure, the event stops looking like a lone associate problem. A firm-level quality-control system existed or was supposed to exist, and it failed somewhere between "draft this" and "file it."

The profession sells hierarchy as a safety mechanism. Junior lawyers draft. Senior lawyers review. Partners sign. Courts and clients are supposed to get something that has survived several rounds of human skepticism. AI shortcuts make that hierarchy look thinner than advertised. If a fabricated citation can move through the chain untouched, the hierarchy has become decorative.

The Broader Alabama Context

The AP story underscored one detail that made the sanctions order more politically loaded: Alabama had paid Butler Snow more than $40 million since 2020 to represent the state in prison-related matters. That number transforms a fake-citation incident from ordinary legal-tech clowning into something taxpayers can understand immediately. Public money was buying premium legal defense, and premium legal defense delivered ChatGPT hallucinations.

The specific lawsuit also concerned prison safety, not some low-consequence contract spat. The allegations involved repeated stabbings and an alleged failure to protect an inmate. Courts tend to become even less patient with procedural nonsense when the underlying claims are that serious. If you are defending a prison system against allegations of violence and neglect, filing made-up authority is about the dumbest possible way to ask a federal judge for trust.

The Vibe Lawyering Pattern

This case fits the now-familiar pattern of vibe lawyering in its purest form. The model outputs something that looks exactly like legal work. The user treats that plausibility as evidence. The normal verification step gets downgraded from mandatory to optional. Then the court does the verification instead.

What makes Butler Snow worth filing in the Vibe Graveyard is not merely that the lawyers used ChatGPT. Lots of lawyers use ChatGPT. What matters is the combination of fake authority, a formal sanctions order, removal from the case, compulsory disclosure across other matters, and a bar referral. That is a full-stack professional failure, not just a bad draft.

If the legal profession wanted a clean example of why "the AI helped with research" is not a defense, it already had several. Butler Snow supplied a more expensive and more humiliating version anyway.