Without a doubt, artificial intelligence tools are helping lawyers do at least some of their work faster than ever before. But what AI giveth, it may also taketh away.
Enter the “verification tax” and “tokenization.” One refers to the time and money users must spend checking the accuracy of AI-produced material. The other reflects the growing, and often unforeseen, costs that firms face as AI providers shift from flat fees to usage-based pricing.
For law firms, in particular, a verification tax may be unavoidable—and expensive. Horror stories continue to multiply about lawyers sanctioned for submitting court documents riddled with AI hallucinations. And the jury is still out on whether legal AI tools deliver the accuracy firms require to trust their outputs without substantial human verification. Indeed, a recent survey commissioned by legal technology provider Morae found that 67% of senior legal professionals feared human verification costs may outweigh the benefits of AI.
Tokenization is giving legal tech managers heartburn as they see AI adoption rates soar and lawyers begin using artificial intelligence to handle a variety of mundane tasks that may end up costing the firms far more than expected. “Law firms bought into AI technology like software and are being billed for it like electricity,” Thomson-Reuters recently noted.
If they get out of hand, either of these developments could scuttle the cost-benefit calculation that many law firms have made around AI. Firms believe that AI tools will help them reduce time spent on routine matters and increase bandwidth for lawyers to take on more premium work. If, however, those lawyers are mired in verifying AI outputs or dealing with inflated token costs, the windfall law firms are anticipating may be severely stunted.
- AI may speed up legal work, but rising “verification tax” and token-based pricing are creating unexpected costs for law firms and legal operations teams.
- Accuracy remains a major risk. Studies found legal AI hallucination rates ranging from under 10% to nearly 50%, while more than 565 U.S. court cases have involved AI-generated errors since 2023.
- Courts and regulators are increasing scrutiny. Lawyers have faced sanctions, fines and disciplinary referrals for submitting filings with fabricated AI content.
- With 85% of firms unable to measure AI ROI, the article examines whether AI’s promised efficiency gains can outweigh mounting verification, compliance and usage expenses.
Cost Concerns: 85% of Firms Can't Measure AI's ROI
How aware firms are of these issues is another question entirely.
AI usage by law firms rose from 28% in 2025 to 41% in 2026, Thomson-Reuters found in a survey released earlier this year. Yet 85% of firms either did not measure or did not know if they measured the return on investment from AI.
That uncertainty is becoming harder to ignore as firms pour ever-more capital into building out their AI capabilities. Earlier this month at ILTACON, the annual conference of the International Legal Technology Association, law firm tech leaders expressed deep concern over token-based pricing that is quickly pushing up the costs for AI usage.
While some at the Nashville event were reporting “impressive individual productivity gains, many professionals acknowledged that demonstrating measurable financial benefits across an entire firm remains difficult,” Thomson-Reuters Institute reported. "Firms are investing millions in this technology, but are we getting that money back?" Littler Knowledge Management Counsel Johnny Wong said.
Law firms have also noticed many lawyers are using the costliest AI tools for routine tasks, “largely because users cannot easily see the underlying token consumption and cost,” Thomson reported. In a profession hardly known for its embrace of the new, legal technologists find themselves in the position of preaching moderation to lawyers. “Lawyers are hooked on this,” one legal tech manager said, according to Thomson.
Studying Legal AI: What the Hallucination Data Actually Shows
One of the key reasons lawyers are using premium AI tools for small tasks is because “they want the AI to be right, and they want the client to get the best,” Thomson said.
The concerns about accuracy are not purely theoretical—or a case of lawyers simply being persnickety. In 2024, Stanford University researchers tested leading legal AI tools that rely on retrieval-augmented generation (RAG). RAG, as Stanford noted, has been “promoted as a solution for reducing hallucinations” by AI-powered legal research products. While they were less prone to hallucination than chatbots like those offered by ChatGPT, they still hallucinated between 17% and 33% of the time.
Presumably, legal AI companies have spent the past few years working to reduce those errors. Indeed, a 2026 study from University of Luxembourg researchers found that the hallucination rate among the best-performing legal AI systems was less than 10%—although the worst-performing platform had a rate of nearly 50%.
In spite of the improvements, the study’s authors wrote that “our results show that hallucinations remain pervasive.” The systems have particular problems with questions based upon a false premise or that contain incorrect assumptions that must be rejected. Those questions “produce high hallucination rates on the manually-drafted questions.”
Lawyers Face Sanctions: The Cases Piling Up in 2026
Meanwhile, the number of lawyers who have been sanctioned by the courts for including hallucinated AI material in court filings continues to climb. On Sept. 14, the New Mexico Supreme Court held Stephen Aarons, a prominent Santa Fe-based criminal defense lawyer, in contempt of court, removed him from a murder case, ordered him to pay $5,000, and referred him to the court’s disciplinary board for submitting a ChatGPT-generated legal brief that included hallucinated testimony from fabricated witnesses.
The court excoriated Aarons for failing to verify the facts and his lack of awareness about AI’s potential shortcomings. “Do you watch the news? Do you listen to the radio? Do you read anything about what’s going on in the world? Because the problem with lawyers relying on AI and hallucinations is an above-the-fold story. Every single day,” Justice C. Shannon Bacon said.
Less than a week earlier, a federal judge in Utah sanctioned two lawyers for Brigham Young University who used a combination of tools—including ClearBrief, Claude, ChatGPT, and Gemini—and submitted filings with a dozen AI-generated errors. In a Sept. 9 order, Judge Ted Stewart of the U.S. District Court for the District of Utah ordered the lawyers to complete two continuing legal education courses on the ethical use of AI in the legal profession within six months. One of the lawyers was fined $2,000 and the other, $1,000.
In his order, Stewart cited the Federal Rules of Civil Procedure’s admonition that lawyers “certify that they have conducted a reasonable inquiry to ensure” that their assertions in court documents are “warranted by existing law.” He added that the lawyers violated the rule and submitted a brief that “was riddled with AI-generated errors.”
Cases Proliferate: 565 Tracked Incidents and Counting
Incidents like the ones in New Mexico and Utah are becoming all too common. A global database maintained by Damien Charlotin, a French lawyer and data scientist, has tracked 565 U.S. court cases between April 2023 and mid-September 2026 where AI used by legal professionals has generated hallucinated content, most often in the form of fake case citations.
The numbers have been escalating. Some 58% of the incidents, or 328 cases, have occurred during 2026. In August alone, 38 new cases were added to the total. And judges are clearly losing patience. The database shows that lawyers have been sanctioned more than 100 times since the beginning of 2025, compared to just twice in 2023 and 2024.
And in all likelihood, the database’s collection of AI-related snafus is only scratching the surface. The database focuses only on filings by courts “where the use of AI, whether established or merely alleged, is addressed in more than a passing reference by the court or tribunal.” And Charlotin cautions that the database does “not track the (necessarily wider) universe of all fake citations or use of AI in court filings.”
In other words, the database is not designed to capture every hallucination going into court documents. Nor is it able to provide insight into possible AI-generated errors in other forms of legal work, such as regulatory filings or documents produced for private businesses and individuals.
Ethical Issues: Why Verification Isn't Optional
As an ethical matter, verification is mandatory. The American Bar Association issued a formal opinion in July 2024 that lawyers using generative AI are ethically obliged to “have a reasonable understanding of the capabilities and limitations” of the technology.
The opinion said lawyers should understand AI well enough to use it competently, protect client confidentiality, supervise its use, and communicate with clients about it. Lawyers who fail to provide “an appropriate degree of independent verification or review of its output could violate the duty to provide competent representation as required by Model Rule 1.1.”
That opinion may also add to the complexity firms face as AI providers shift their pricing models. When a third-party AI provider charges a lawyer per use, “it would ordinarily be reasonable for the lawyer to bill the client as an expense for the actual out-of-pocket expense incurred for using that tool,” the opinion said. But if AI tools function as a general law firm technology, they should be treated as overhead that cannot be passed on as a direct legal expense.
As token-based pricing spreads, firms may shift some matter-specific costs onto their clients. But unless they reach an advance agreement with a client, lawyers may not mark up their services to turn technology expenses into “an additional source of profit for the law firm.” In other words, token-based pricing may create higher AI bills and new accounting and billing questions about who ultimately pays the tab.
A Key Question
Law firms must balance the cost of using AI and the expense of ensuring its work is accurate. As AI models develop, they may help firms bring down both of those bills.
This provokes a fundamental question: After law firms pay for the software, the rising token-related expenses, and the work needed to verify outputs, how much will they actually save? Given most firms don't have a clear idea of AI’s ROI, the answer at this point is “no one really knows.”
---
David L. Brown is a legal affairs writer and consultant, who has served as head of editorial at ALM Media, editor-in-chief of The National Law Journal and Legal Times, and executive editor of The American Lawyer. He consults on thought leadership strategy and creates in-depth content for legal industry clients and works closely with Best Law Firms as senior content consultant.