AI Is a Tool, Not an Attorney: How Clients’ Use of AI Can Cause Them to Pay More in Legal Fees
Artificial Intelligence is on the precipice of ubiquity. It is a tool with seemingly infinite uses including helping us write “Thank You” letters, find cooking recipes, and even creating fun pictures based on a text prompt. For all intents and purposes, AI has been extremely helpful, even fun. Accordingly, it is only natural for people to turn to AI to help with their legal questions with the added benefits of being an inexpensive and instant. But “inexpensive” and “instant” do not equate to “good” or even “correct.” In fact, using AI for legal questions, drafting contracts, or performing case research may ultimately cost clients significantly more money in fees or adverse monetary judgments in the long run.
AI Is Built On An Unsecured Foundation
Pulling back the curtain of the AI prompt screen, is a program built upon algorithms to imitate human behavior based on inaccurate, or at best, outdated legal information. At worst, AI will completely make up legal authorities called “hallucinations.”1
AI systems rely on data inputs, aka prompts, and if these inputs are flawed, the outputs will undoubtedly be as well. Many AI tools have been found to perpetuate biases present in their training data, potentially leading to misleading results.2 Accordingly, if clients rely on AI-generated analysis without attorney oversight, then there is a strong possibility your client could rely on incorrect legal citations or misinterpretations of case law, ultimately, jeopardizing their position.
Further, these biases are not just present in AI itself. When clients use AI for legal questions, they phrase prompts in a way that reflect their desired outcome rather than neutrally. Their own inherit bias creates a detrimental problem because AI tools tend to tell its users what they want to hear. The models are designed to be helpful, and they often provide confident sounding but legally inaccurate or misleading responses tailored to the assumptions in the prompt.
Moreover, the law is not static. Laws change and they can be interpreted differently among various jurisdictions, and in some cases, even courts within the same jurisdiction. This presents a significant issue because AI that is trained on older data, or worse, provide responses inapplicable to a client’s specific jurisdiction. Thus, AI will miss recent court rulings, new statutes or state-specific rules.
Another issue with AI is, and this may come as a surprise, that it is not human. AI has no wisdom or experience. The practice of law requires the application of a unique set of facts to the law. Most AI tools are not trained to weigh client-specific factors with risks and strategic considerations — all necessary aspects of the art of practicing law. To put more eloquently, I’ll rely on Oliver Wendell Holmes’ famous quote, “The life of the law has not been logic; it has been experience.” So, despite analyzing terabytes of data to train its models, AI has no experience, and ultimately, its use may cause the client more harm than help.
The Costs Clients Incur From Using AI As Their Attorney
It is not uncommon for clients to bring their “research” to their attorneys expecting confirmation, slight refinement, or further explanation. Instead, the attorney must analyze every argument and fact-check every cited case to ensure it is not a hallucination. Frequently, attorneys will have to take even more time to explain to the client why the AI arguments or claims are not tenable. In other words, attorneys will need to spend more time than if they had been simply consulted from the start.
Another common example involves a client using AI to draft a contract. Although the contract may appear like a well-written contract with verbose “legalese,” these often miss critical legal terms or include outdated, unenforceable provisions. Such faulty contracts can hurt clients infinitely more than had they just used an attorney to draft contracts tailored for them instead of a full redraft of AI slop.
Importantly a client’s use of AI does not just hit them in the wallet, its use may need to be disclosed in discovery. Courts have found AI use for legal matters is not legally protected. The information a client used in a prompt and the AI produced in response is all discoverable.3 In U.S. v. Heppner a trial court found client AI prompts and generated documents were not privileged because by inputting confidential information into a consumer AI platform operated by a third party, the client voluntarily disclosed that information outside the attorney-client relationship. Moreover, Heppner noted the AI platform’s terms of service and privacy policy that permit data collection, retention, and found that these policies negated any reasonable expectation of confidentiality.
Thus, it is critical that attorneys stress the detrimental effects their AI use may have on their case.
What To Do
This article is not intended to cast aspersions on the use of AI in the legal field. It is an incredibly powerful tool getting more and more valuable every passing year. AI can review documents, summarize depositions, and generate outlines of cases in the snap of a finger. However, it still lacks the practical and legal knowledge of an attorney to tailor specific facts in a specific jurisdiction to determine the best course of action for a client.
At the outset of representation clients must understand that AI is unreliable when it comes to (1) legal research, (2) jurisdictional distinction, and (3) practical, common sense judgment. Moreover, if a client is to use AI for their legal issues, they must make sure that the AI platform they use (1) does not retain data or use it for data training; (2) uses confidentiality protections for input data; and (3) use AI under the attorney’s direction and supervision as part of the attorney’s legal strategy. Understanding each of these points will help save your client money, time, and energy.
—
1 Stanford University Human-Centered Artificial Intelligence: Dan Ho, et al.: “AI on Trial: Legal Models Hallucinate in 1 out of 6 (or More) Benchmarking Queries” (evaluating claims by LexisNexis (creator of Lexis+ AI) and Thomson Reuters (creator of Westlaw AI-Assisted Research and Ask Practical Law AI) that their use of retrieval-augmented generation (RAG) helps significantly “‘avoid’ hallucinations and guarantee ‘hallucination-free’ legal citations.” The study found that , while RAG systems “do reduce errors compared to general-purpose AI models like GPT-4 [–] a substantial improvement [–] these bespoke legal AI tools still hallucinate an alarming amount of the time: the Lexis+ AI and Ask Practical Law AI systems produced incorrect information more than 17% of the time, while Westlaw’s AI-Assisted Research hallucinated more than 34% of the time.”)
2 LLM models used in healthcare “may recapitulate harmful, race-based . . . content when responding to eight different scenarios that check for race-based medicine or widespread misconceptions around race. . . . This study shows that based on our findings, these LLMs could potentially cause harm by perpetuating debunked, racist ideas.” (Politico, “Why Your Chatbot’s So Racist,” Dec. 12, 2023 (report on study showing racial bias in AI models used in healthcare).)
3 United States of America v. Bradley Heppner (S.D.N.Y. 2026) 25 Cr. 503. There, a former CEO accused of swindling investors out of $300 million. After the CEO received a grand jury subpoena, he used an AI platform by typing details he got from his lawyers into the AI’s command prompt. The AI tool generated numerous documents in response to the CEO’s prompts, including an outline of a defense strategy based on the charges his lawyers anticipated. The government argued such documents were discoverable. The CEO argued they were not because the documents were protected by the attorney-client privilege that applies to third parties such as a consulting witness; attorney-client privilege; and work-product doctrine.