Generative AI is reshaping legal services at remarkable speed, and law firms are adopting it to deliver faster, more efficient results for clients. But efficiency does not shift professional responsibility from the lawyer to the technology. Every attorney who uses AI must diligently review and verify its output.
Most people have heard of lawyers who filed documents containing AI-“hallucinated” authorities. But AI’s risks extend beyond hallucinations and beyond litigation. AI tools are only as good as their inputs, so limited, outdated or incomplete sources yield unreliable results. A tool trained on outdated law or contracts, for instance, may draft unenforceable provisions. AI can also combine accurate information inaccurately, such as by misreading how complex regulations and statutes interact. And AI tools often lack the commercial context that prudent client advice requires.
Diligent verification is therefore an essential best practice, not an afterthought. Attorneys who fail to verify AI output risk violating ethical rules, harming their clients and incurring sanctions. And, as the rules and incidents below show, a written policy alone will not prevent improper AI use.
The Ethical Rules and Various Court Rules Require Sufficient Verification
On the ethical side, the American Bar Association issued a formal opinion addressing how the Model Rules of Professional Conduct apply to AI use. The opinion cautions that the duty of “competent representation” typically requires lawyers to verify AI-generated output before relying on it. According to the ABA, “[t]he appropriate amount of verification ... will necessarily depend” on the task and the tool: “less verification” may suffice for “generat[ing] ideas,” but tasks that require personal judgment — offering legal advice, negotiating claims and making representations to a court — demand more.
Some court systems have gone further and adopted rules that formalize a verification requirement. The Florida Supreme Court, for example, recently amended a rule to require lawyers in Florida courts to verify that legal authorities cited in filings “exist and are accurately cited,” and authorized courts to sanction attorneys who fail to do so. To prevent a patchwork of obligations among lower courts, the Florida Supreme Court then issued an administrative order barring those courts from imposing their own AI disclosure or certification requirements.
The New York Unified Court System, by contrast, issued a model rule that, if individual courts adopt it, would require attorneys using AI to “carefully review” a filing, “independently ensure that [it] contains no fabricated or fictitious cases, statutes, or other material,” and certify as much. The model rule also authorizes sanctions for violations. Unlike Florida courts, however, New York courts remain free to adopt the rule or craft their own AI-use requirements.
In federal court, failing to verify filings may violate local rules, a judge’s individual practices or procedural rules such as Rule 11 of the Federal Rules of Civil Procedure, which requires attorneys to certify that factual contentions have evidentiary support and that legal contentions are warranted by existing law. See Mata v. Avianca Inc., 678 F. Supp. 3d 443, 461 (S.D.N.Y. 2023).
A verification failure can also trigger a court’s inherent disciplinary authority, as a recent 6th U.S. Circuit Court of Appeals order suggests. In United States v. Kuhl, the 6th Circuit ordered an attorney to show cause why the court should not sanction or discipline him under the Federal Rules of Appellate Procedure and its inherent authority, citing two apparently hallucinated cases and numerous inaccurate statements of law. The attorney responded that an insufficiently rigorous verification practice caused the errors but asked the court not to impose discipline because he did not intend to mislead. The matter remains pending as of this writing.
Because the landscape is shifting quickly, counsel practicing across jurisdictions should confirm they comply with each governing rule. Some jurisdictions impose obligations beyond verifying AI-generated statements of law and fact, including certifications or disclosures regarding AI use.
Independent Verification Is Key
Many organizations have adopted AI policies, but a policy alone does not ensure compliance. The examples below show why independent verification by the responsible attorney remains essential.
One large, well-regarded firm offers a cautionary tale. Its written policy required lawyers to independently check AI-generated answers, case citations and other work product, summed up in a simple instruction: “Trust nothing and verify everything.” Despite that policy, the firm had to apologize to a court in writing after opposing counsel found numerous AI-generated errors in one of its motions.
Johnson v. Dunn, 792 F. Supp. 3d 1241 (N.D. Ala. 2025), involved a similar incident with severe consequences for the attorneys involved. Their firm required prior approval to use AI for legal research and independent verification of its output through traditional methods. Despite that policy, one attorney used generative AI without permission and without verifying the citations, and the resulting filings contained fabricated legal authorities. The court (1) reprimanded the attorney who used the AI tool, the attorney who signed the filing and the supervising attorney; (2) disqualified all three, even though two never used the AI tool themselves; and (3) referred all three to licensing authorities.
In Mezu v. Mezu, 267 Md. App. 354 (2025), the court reaffirmed that lawyers cannot escape their verification duties because someone else used AI. A nonlawyer clerk used ChatGPT to find cases for an appellate brief, then used it again to check the citations. The brief contained nonexistent cases, misquotations and real cases that did not support the propositions cited. Counsel said he had never personally used AI and had relied on the clerk’s assurance that the authorities were verified. That reliance did not relieve him of responsibility for the brief he signed, and the court referred him to the Attorney Grievance Commission.
Mezu also highlights a key point: Attorneys should not use AI to verify AI’s own work. Asking an AI tool whether a citation is correct is no substitute for reading the underlying authority.
When a verification failure occurs, the response matters. In Mata, the court sanctioned the attorneys in part because they stood behind fictitious authorities even after opposing counsel and the court questioned whether the cases existed, and the court observed that prompt candor would have changed the outcome. Courts generally weigh remedial conduct, such as prompt withdrawal of the filing, disclosure and corrective measures, when deciding on sanctions.
Key Takeaways
Technology will keep changing, but the fundamental obligation will not: The lawyer whose name is on the work remains responsible for getting it right. As a best practice, lawyers should independently confirm substantive AI-generated work before relying on it, and that responsibility extends to AI use by junior lawyers, clerks and anyone else under an attorney’s supervision. AI can be wrong, so verify every output before relying on it in any significant way.
