AI in Summer 2025: The Promise, Potential, and Pitfalls

Recently, our firm evaluated the Artificial Intelligence (AI) assisted tools offered by Lexis and Thomson Reuters Westlaw and the attorneys have had opportunity to test-drive both. The results can be, like ChatGPT, extraordinary. The tools can help draft contract provisions in seconds, condense hundreds of pages into a few tight paragraphs, and suggest creative ways to shape arguments. The benefits are obvious — fewer hours working with a form or template, more time for focus on analysis or strategic judgment to advance a client’s case or matter. That is the promise, and these tools do offer opportunities for creating significant efficiencies for practitioners that simply were not available even just a few years ago.

The pitfalls, however, are just as clear. These models are designed to give an answer — and not necessarily the correct one. Since ChatGPT’s public release in November 2022, the technology has advanced rapidly, and the current version, ChatGPT-5, is far more capable than its earliest iterations. OpenAI touts that GPT-5 is their “smartest, fastest, most useful model yet, with built-in thinking that puts expert-level intelligence in everyone’s hands.” However, the problem of hallucinations remains. As sophisticated as generative AI tools are becoming, they may still make things up. For lawyers, those hallucinations can take the form of nonexistent citations, quotations that never appeared in any reporter, misleading characterizations of authority, or entire legal doctrines conjured out of thin air.

There have been some recent decisions which serve as reminders of the potential pitfalls practitioners face when they fail to check or double check the sources and all citation to authority generated using an AI tool. The most popular AI chatbot is currently ChatGPT with roughly 350 million monthly users (as of January 2025). Although the AI tool has seen multiple advancements and the current version is designed to be helpful and provide the desired answer, there continue to be regular reports of non-existent authority (e.g. fake cases, misleading quotes, self-serving interpretations of actual authority, or outright made-up legal principles). The appropriate response to the query of, “Should we submit hallucinated cases or make-believe authority in briefs, memoranda, or other court filings?” should be an emphatic, “Never.” And the reader is encouraged to adopt this philosophy and otherwise comply with Rule 11 of the Federal Rule of Civil Procedure (FRCP) in all respects. Still, in several cases decided in the last ten months, multiple attorneys fell short. Their situations offer some timeless lessons. 

In Shahid v. Esaam, 2025 WL 1792657 (Georgia, June 30, 2025), a divorce matter, the Georgia Court of Appeals granted the wife-appellant’s application for discretionary review and vacated the lower court’s order, after determining the appellate brief completely ignored wife's argument that trial court's order contained two fictitious case citations, and that counsel had provided 11 bogus case citations out of 15 total. The appeals court wrote “we are troubled by the citation of bogus cases in the trial court's order.”

In Coomer v. Lindell, 2025 WL 1865282 (Colorado, July 7, 2025), the federal court issued 4-figure sanctions against one lawyer and his firm, jointly and severally, and a second lawyer individually for Rule 11 violations in the form of submitting briefs that contained hallucinated cases and the lawyer admitted to not double-checking any of the citations prior to submission. 

Most recently, in Johnson v. Dunn, (Case No. 2:21-cv-1701, July 23, 2025), the U.S. District Court for the Northern District of Alabama evaluated claims regarding fabricated citations to legal authorities and disqualified three lawyers from the case after determining that they had included citations hallucinated by ChatGPT in motion filings. The court found the three lawyers benefitted from repeated warnings, internal controls, and firm policies about the dangers of AI misuse and that they have regular access to gold-standard legal research databases. In addition to being disqualified from further participation in the case, the judge publicly reprimanded the attorneys, and the matter was referred to the Alabama State Bar and other applicable licensing authorities.

The takeaways should be clear: Do the work. Do good work. Do all of the work. One of my favorite law professors often told our class to “read the statute or the contract” and if you do, “you will save or earn your clients lots of money.” Diligence and competency are the cornerstones of good to exceptional advocacy. For the practitioners in 2025, utilize tools that will support efficiencies in your practice and that will benefit your clients but at the same time, one must still read the cases. If you do, you will enjoy avoiding the discomfort of scrambling to respond to sanctions, motions, or inquiries like in the Lindell and Shahid matters or being deeply embarrassed like the lawyers in Alabama.  Most notably, your clients, your colleagues, and the tribunal will all appreciate knowing that the authority cited in your brief or related submission is rock solid and reliable.  

David M. Tang is a Partner at Underberg & Kessler LLP and serves as Chair of the firm’s Health Care and Creditor’s Rights practice groups. He advises clients on business, health care, restructuring, and commercial litigation matters. David can be reached at dtang@underbergkessler.com  or 585.258.2845.

Reprinted with permission from The Daily Record and available as a PDF file here.