Did AI Make the Mistake, or Did the Lawyer? U.S. Courts Are Taking a Stricter Approach to Lawyers' Use of AI

In 2023, Mata v. Avianca, Inc. brought the professional risks of generative AI to the attention of the legal profession. Counsel used ChatGPT while preparing a court filing and submitted several nonexistent cases without verifying them. The U.S. District Court for the Southern District of New York sanctioned two of the offending lawyers and their law firm. The decision quickly became a leading cautionary example in discussions of AI-assisted legal practice. See Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023).

More than two years later, nearly every lawyer has heard that story. At the same time, generative AI has entered ordinary legal practice. Lawyers now use it for legal research, contract drafting, document review, evidence organization, case analysis, and preliminary drafting. For many lawyers, AI has become an important tool for improving efficiency.

A new question has therefore emerged: once the profession knows that generative AI can fabricate authorities, will a court still accept "I did not know AI could be wrong" as an excuse? In June 2026, the U.S. District Court for the Northern District of Mississippi gave a clear answer in Withers v. City of Aberdeen, No. 1:24-CV-218-SA-RP, 2026 U.S. Dist. LEXIS 126060 (N.D. Miss. June 8, 2026).

One Case, Four Lawyers, and Sanctions on Both Sides

Unlike Mata, Withers did not involve a single lawyer's misuse of AI. Filings submitted for both the plaintiff and the defendant cited cases that did not exist. The affected filings also bore the electronic signatures of both the out-of-state lawyers and their local counsel.

The court issued an order to show cause. The two out-of-state lawyers admitted that they had used AI-generated material without independently verifying the cited authorities. Their local counsel admitted that they had not reviewed the filings, even though their electronic signatures appeared on them. The resulting sanctions were severe.

The court revoked the pro hac vice admissions of the two out-of-state lawyers, barred them from appearing in the Northern District of Mississippi for two years, and fined them $2,500 and $3,500. It also required one of them to complete a continuing legal education course on AI and legal ethics. The two local lawyers, who had not used AI or drafted the filings, were disqualified from further participation in the case and fined $1,000 each. The court also directed the clerk to send the sanctions order to the Mississippi, Louisiana, and Texas bars.

The Court Did Not Prohibit the Use of AI

These sanctions may appear to reflect judicial hostility toward AI, but the reality is more nuanced.

Judge Sharion Aycock expressly recognized that AI can be a powerful tool when used prudently. Quoting prior decisions, the court also stressed that generative technology can produce words, but it cannot supply truth, sincerity, or responsibility. Those duties remain with the lawyer who signs the filing.

The court's message was straightforward: AI may assist with drafting, but it cannot assume the lawyer's duties of candor, verification, and professional responsibility.

This reasoning reflects the direction of recent U.S. decisions. Courts are not rejecting AI as a legal tool, and they are not ordering lawyers to stop using it. The problem arises when a lawyer transfers professional judgment to the tool and files its output without adequate review.

The most important difference between Mata and Withers is not simply the severity of the sanctions. It is the court's treatment of claimed ignorance. Relying on a recent Fifth Circuit decision, the court concluded that ignorance of the risks of generative AI is no longer a credible excuse.

Even if ignorance of those risks might once have been an excuse, it is not one now.

The court reasoned that repeated incidents involving AI-generated false citations have made the risks well established. A lawyer who uses AI must therefore understand that it can hallucinate and must use a reliable process to verify every authority and every material factual assertion.

The court also repeated a rule that every lawyer should take seriously:

Quoting Mattox v. Product Innovations Research, LLC, the court stated that the Rule 11 obligation "is absolute. It cannot be outsourced to technology or delegated to co-counsel."

This does not make Rule 11 a strict-liability rule for every mistake. It does mean that each lawyer who signs, files, submits, or later advocates a paper must conduct an inquiry reasonable under the circumstances. A lawyer cannot avoid that duty by blaming an AI tool, lead counsel, or local counsel.

In the AI Era, Lawyers Need More Than Prompting Skills

For the past several years, many people have asked whether AI will replace lawyers. Recent decisions point to a different question:

AI may not replace lawyers, but a lawyer's failure to use AI responsibly may call the lawyer's competence into question.

AI can perform an increasing number of repetitive tasks. It cannot assume a lawyer's professional obligations or duty of candor to the tribunal. From Mata to Withers, the judicial message has become clearer: lawyers must learn not only how to use AI, but also how to verify its output.

For the legal profession, the real distinction may no longer be between lawyers who use AI and lawyers who do not. It may be between lawyers who gain efficiency from AI while preserving independent judgment, careful verification, and personal accountability, and those who do not. AI does not reduce a lawyer's responsibilities. It raises the level of care that courts expect. Technology may improve efficiency, but legal judgment, factual verification, and professional responsibility cannot be outsourced. That is also the principle we follow when using AI tools in cross-border litigation and commercial disputes.