When AI does the lawyering and ethics take a back seat

"We've got a long way to go and a short time to get there," a lyric from the 1977 comedy Smokey and the Bandit perfectly captures how many attorneys feel every day. With the deadline for an important legal brief fast approaching, the workload far exceeds the time available, and no one on the litigation team has time to help. What to do?

Increasingly, attorneys are turning to Generative AI as a substitute rather than a tool. After all, AI seems authoritative, coherent and confident. But mostly, it's fast. So, the attorney asks AI to provide a summary of the five best cases on a legal issue and AI delivers a polished paragraph of cases with pinpoint cites and quotes in a compelling format and under a tight deadline.

The attorney turns the AI-generated summary into a paragraph for the brief and files it with the court without verifying that the cited authorities exist and are still good law, that propositions match holdings, or that quotations are accurate. Blind reliance on AI not only causes embarrassment and sanctions but can also foster dependency or exacerbate mental health issues.

A group of researchers recently documented real examples of people who began to experience mental health issues after intensive use of AI systems. "Delusions by design? How everyday AIs might be fuelling [sic] psychosis (and what can be done about it)," Hamilton Morrin et al. The researchers found that AI's ability to mimic human conversation may cause the user to form an attachment to an apparently sentient AI entity. The researchers found a common pattern. AI is used for assistance with everyday tasks, which builds trust and familiarity with the system. Because AI systems are designed to maximize engagement with the user, it may create a "slippery slope" in which the user becomes increasingly dependent on the AI system and more unmoored from "consensus reality."

It's easy to see how a harried attorney might rely too much on an AI system that reinforces his or her theory of a case or that quickly provides cases that open a previously unconsidered line of attack or defense --if the cases are real and still good law. The appeal of using AI is that the platforms are designed to maximize the chance of giving an answer, which is welcome if an attorney is up against an important deadline. But the motivation to come up with an answer, any answer, means that the AI platform may make up an answer that seems plausible rather than admit it doesn't know something.

This is not a hypothetical problem. Examples of attorneys being sanctioned for AI hallucinations in court briefs are on the rise. Last fall, a California appellate court noted that in the last two years, many courts have confronted briefs populated with fraudulent legal citations resulting from attorneys' reliance on AI. Noland v. Land of the Free, L.P. (2025), 114 Cal. App. 5th 426, 443-44. "The appearance of hallucinated citations in briefs generated from AI is no longer in its nascent stage. Regrettably, the number and regularity with which courts have been faced with hallucinations in court filings continues to rise." (Id.) The court lamented that "the problem of AI hallucinations is getting worse, not better, noting that OpenAI's newest models hallucinated "30-50% of the time, according to company tests." (Noland, citing (Murray, "Why AI 'Hallucinations' Are Worse Than Ever," Forbes.com (May 6, 2025)).

The Noland court stated that "hallucination" is a particularly apt word to describe "the darker consequences of AI" in which there are many instances ... where hallucinations are circulated, believed, and become 'fact' and 'law' in some minds." (Id.) The court noted that submitting legal documents with fake cites "is entirely preventable by competent counsel who do their jobs properly and competently." (Id.) In Noland, the attorney filed an appeal that had numerous fake cites, but once discovered by the court, he claimed he was unaware that AI platforms could fabricate legal authority and had no intention of misleading the court. The court noted the numerous articles and other court decisions discussing AI hallucinations and sanctioned the attorney $10,000.

Becoming too dependent on Generative AI is risky. What happens if the platform crashes or a communications outage prevents an attorney from accessing the AI platform to complete a pleading that is now so late there is no other option than to use AI? Copying and pasting AI cites or passages into a pleading raises a significant likelihood of professional embarrassment (at the very least), but also malpractice claims from affected clients or sanctions.

California's sanctions regime for improper filings reflects a policy judgment that courts must be protected from irresponsible paper practice. Code of Civil Procedure section 128.7 authorizes trial courts to impose sanctions to check abuses in the filing of pleadings, petitions, notices of motion and similar papers. Section 128.7 contemplates sanctions against attorneys and law firms responsible for the violation and provides that, absent exceptional circumstances, a law firm is jointly responsible for violations committed by its lawyers and employees. That provision creates special risk for firms where AI use is informal, untrained or unsupervised within a firm.

Attorneys need to be mindful that AI is a powerful tool, not a substitute for diligent legal work. The fact that many attorneys feel so pressed that they knowingly copy someone else's work (whether produced by AI or other attorneys) signal problems within the profession that need to be addressed.