Don't Bet the Business

On June 17, the Michigan Court of Appeals handed down an opinion in Barber v. Morawa that every attorney using AI tools needs to read.

The case started as a fairly run-of-the-mill medical malpractice dispute, but it turned into a landmark ruling on what happens when lawyers submit AI-generated legal citations without verifying them. Let's break it down.

The Facts: A Hip Replacement Gone Wrong (Allegedly)

The underlying case involves Joyce Barber, who sued Dr. Lawrence J. Morawa, an orthopedic surgeon, over a hip replacement surgery. After a five-day trial, the jury returned a verdict finding that Dr. Morawa was not negligent. 

When the court polled the jurors, five confirmed the verdict, but the foreperson—Juror 7—said it did not accurately reflect his vote. Because only five agreeing jurors are needed in a Michigan civil case, the verdict stood.  

After the verdict, an off-the-record discussion took place between the attorneys, the judge and the jurors. After that, Barber moved for a new trial or evidentiary hearing, claiming the verdict was tainted by juror misconduct. Specifically, she alleged that one juror admitted to consulting his mother (a hip surgery patient) about X-rays and the case; another juror admitted to conducting independent online research about a 25-millimeter screw; the foreperson was prevented from reviewing trial exhibits; and alternate jurors stayed in the jury room during deliberations.

The defendant disputed each of those characterizations. According to defense counsel's notarized affidavits, no juror consulted a family member about case evidence, no one did independent research, and the foreperson voluntarily chose not to hold up the vote because he was already "90% for the defense." The trial court sided with the defendant's account, denied the motion, and plaintiff appealed.

The AI Problem: Fake Cases, Fake Quotes, Repeat Offenses

Now, here's the part that made this case noteworthy. Throughout the proceedings—both at the trial court level and on appeal—plaintiff's counsel repeatedly cited legal authorities that simply did not exist. This started in December 2024 with a motion for a protective order that cited "People v. O'Keefe, 2019 WL 6907273"—a case that does not exist. The defendant flagged it, but counsel never corrected it.

Two months later, in the motion for a new trial, plaintiff’s counsel cited two more nonexistent criminal cases: "People v. Stubblefield" and "People v. Royster." Again, the defendant identified the problem. In a reply, plaintiff’s counsel dismissed it as "an innocent and inadvertent mistake."

On appeal, the pattern continued. Plaintiff’s counsel cited "Meyer v. Walker, 241 Mich. App. 295 (2000)"—another fabricated case. He also cited real cases like People v. Budzyn and People v. Miller for propositions those cases never actually stated. When the defense pointed all of this out in its responsive brief, counsel filed a reply brief without even acknowledging the problems.

Eight months later—just weeks before oral argument—counsel filed a "Notice of Correction." In it, he accepted "full responsibility," attributed the errors to "over-reliance on artificial intelligence research tools" that generated "plausible but fabricated case citations—a known limitation of such tools," and said he had "implemented verification protocols to prevent recurrence."

But here's the kicker. The notice of correction, itself, was prepared with the help of AI, and it still attributed quotations and propositions to cases that didn't contain them. Counsel admitted this at oral argument.

Court of Appeals Ruling

The appellate court addressed two main questions. First, was the trial court right in denying the new trial motion? The appellate court  found that plaintiff's motion depended on facts outside the record that were not supported by valid affidavits as required by MCR 2.611(D)(1). Even setting aside that procedural defect, the appellate court found that the plaintiff failed to show any alleged misconduct materially affecting her substantial rights, relying instead on criminal-law presumptions inapplicable to a civil malpractice action. The appellate court affirmed, holding the trial court did not abuse its discretion in denying the plaintiff a new trial.

What About the Fabricated Citations?

This is where the appellate court broke new ground.  The panel held that plaintiff’s counsel's repeated submission of fabricated and unsupported authority violated the duty of reasonable inquiry under MCR 1.109(E)(5). Under this rule, an attorney's signature on a filing certifies that the document "is warranted by existing law" after "reasonable inquiry."  The reasonableness standard is objective—subjective good faith is irrelevant.

The appellate court joined federal jurisdictions in holding that relying on generative AI without meaningful verification violates this duty. The appellate court emphasized that "artificial intelligence may be a useful tool for legal research and drafting, but the use of such technology does not alter an attorney's professional obligations." Lawyers must verify that cited authorities exist, actually read the cases they cite and ensure those cases support the propositions asserted.

The appellate court then found sanctions appropriate as authorized under both MCR 7.216(C)(1)(b) (for vexatious appellate proceedings) and MCR 1.109(E)(6) (mandatory sanctions for signature-rule violations). It remanded for the trial court to determine the actual damages and reasonable attorney fees defendant incurred because of the appeal, payable by plaintiff's counsel personally. The appellate court also directed the its clerk to forward the opinion to the Michigan Attorney Grievance Commission for possible investigation.

Why This Case Matters

This is the first published Michigan appellate decision to address sanctions for AI-generated fabricated legal authority. While federal courts have been sanctioning lawyers for AI hallucinations for a few years now under Rule 11—from Mata v. Avianca in 2023 to the U.S. Court of Appeals for the Sixth Circuit's decision in Whiting v. City of Athens in 2026—Michigan state courts had not yet weighed in. Barber v. Morawa fills that gap.

A few key takeaways for practitioners. First, the court made clear that the "I didn't know AI could do that" excuse has a very short shelf life. Counsel here was warned multiple times that his filings contained fake citations, and yet he kept submitting AI-assisted work without adequate verification. The court treated this pattern as demonstrating "repeated carelessness" rather than an isolated mistake.

Second, the sanctions imposed were personal. The damages are payable by counsel, not the client. That's a significant deterrent. And the referral to the state’s attorney grievance commission signals that disciplinary consequences could follow.

Third, the court's framing is worth noting. It did not ban AI use or suggest that AI tools are inherently problematic. It simply held that using AI doesn't change a lawyer's existing obligations. You must still read the cases cited. You must still verify they exist. You must still confirm they say what you think they say. AI technology is new, but an attorney’s duty isn't.

For Michigan practitioners—and really, for any attorney using AI research tools—the message is clear: verify everything. AI is a starting point, not an endpoint. And if someone tells you your citations are fake, stop using the same unverified AI process and fix the problem immediately.