The genre of "lawyer lets a chatbot write the brief and a federal judge notices" is now so well established that the only thing left to track is the body count. Add one more. On June 22, 2026, Chief Judge Matthew W. Brann of the U.S. District Court for the Middle District of Pennsylvania suspended attorney Nicholas W. Mattiacci Sr. of Jenkintown for six months and ordered him to pay a $1,500 sanction, after a post-trial brief he filed turned out to be held together with citations that do not exist.

A serious case, a careless filing

The dispute underneath the sanction is not trivial. Mattiacci represented a plaintiff, a former Montoursville Area High School baseball coach who lost an eye after being struck by a batted ball, in a civil suit against equipment makers including BSN Sports and Garware Technical Fibres. That is exactly the kind of case where the briefing is supposed to be airtight, because a real person with a real injury is depending on it.

Instead, the post-trial motion and brief came in carrying fabricated case citations, quotations that do not appear in the cases they were attributed to, and irrelevant authority dropped in as if it were on point. According to the court's accounting, there were fewer than ten correct case citations in the filing. The Rules of Civil Procedure ask for something almost insultingly basic here: read your documents before you file them, and make sure the cases you cite are real and say what you claim. That bar was not cleared.

Brann was less bothered by the AI than by the excuses

In a May 21 opinion that set up the sanction, Brann determined the hallucinated citations violated Federal Rule of Civil Procedure 11, the long-standing rule that says when you sign a filing you are certifying you actually checked it. What clearly irritated the court was not just the machine error; it was the response to being caught.

Brann rejected the attempt to hide behind the software. As he put it, a lawyer cannot escape the duties of his role "by pointing to the duties of others, ascribing a label to his errors, or by claiming that pointing out his errors are attempts to misdirect." The judge described Mattiacci's reply brief as including a long soliloquy on the hardships of solo practitioners, an argument that the citation errors were minor, accusations about opposing counsel, and a characterization of the whole problem as "procedural distractions." None of that landed. Courts have heard the "my research tool did it" defense enough times now that it reads less like an explanation and more like a confession that nobody verified the work.

The penalty

The order suspended Mattiacci from practicing in the Middle District of Pennsylvania for six months, effective June 22, and imposed the $1,500 fine. A suspension is a heavier consequence than the cash penalties that defined the earlier wave of AI-citation cases, and that escalation is the point. When the first ChatGPT-citation sanctions hit in 2023, courts treated them as a novel embarrassment and reached for modest fines. By mid-2026, judges have run out of patience for novelty. The behavior is well understood, the warnings are everywhere, and the tariff is going up.

This was not his first time

The detail that turns this from an unlucky one-off into a pattern: Mattiacci had already been sanctioned by the New Jersey Superior Court for similar conduct in 2025. The defendants in the Pennsylvania matter were also seeking their own sanctions against him. Doing this once, apologizing, and fixing your process is a story about a tool that fooled a busy practitioner. Doing it again after being formally disciplined for it is a story about a workflow that never actually changed.

The apology

To his credit, Mattiacci's public statement leaned into contrition rather than spin. He said the episode was "deeply troubling," offered "no excuses, no attempts to deflect," and accepted "full, personal responsibility for the breakdown in my post-trial verification of cited case law." He acknowledged that he "implemented this new software into my practice without fully understanding its function and capabilities," and said he had "permanently overhauled my practice protocols to ensure absolute manual verification of all case law moving forward."

That last line is the whole lesson compressed into a sentence. The failure was never that he used an AI tool. The failure was treating its output as finished legal research rather than as an unverified draft that happens to look authoritative. A generative model will hand you a case name, a reporter citation, a pinpoint, and a confident quotation, and every one of those can be invented. The text looks exactly like real law because looking like real law is the only thing the model is actually good at.

Why this keeps happening

The uncomfortable truth is that nothing about this requires a new rule. Rule 11 already said you have to verify your filings. Professional conduct rules already required competence. The AI did not create a loophole; it just industrialized an old failure mode by making it trivial to generate large volumes of plausible, wrong citations faster than a tired solo practitioner can check them. The tool lowered the effort of producing legal text to nearly zero while leaving the effort of verifying it exactly where it always was.

For everyone watching from the sidelines, the takeaway is narrow and boring and correct: if you cannot pull a cited case up in a real legal database and confirm it says what your brief claims, that citation does not go in the brief. A chatbot that cannot show its work through real sources is not doing legal research. It is doing a convincing impression of legal research, and the difference is the part that gets you suspended.