Why companies may be liable for AI errors

A U.S. Tax Court decision in Clinco v. Commissioner found that three of four case citations in an attorney's brief appeared to be AI-generated hallucinations referencing nonexistent Tax Court cases, prompting the court to reject the taxpayer's argument and warn attorney Abraham Wagner of possible sanctions, according to an analysis by two certified public accountants published in the September 2026 issue of Journal of Accountancy. The case involved a restaurant owner accused of underreporting more than $2 million in business receipts and claiming unsupported depreciation deductions on rental properties. The authors cite the ruling alongside Deloitte's 2025 refund to the Australian government over a 237-page report containing fabricated citations and a nonexistent quotation to argue that professionals, not AI developers, bear liability for AI-generated errors.

Artificial intelligence https://www.fastcompany.com/section/artificial-intelligence has moved quickly from a novelty on the job to a common workplace tool. Gallup found that 52% of U.S. employees https://www.gallup.com/workplace/704225/rising-adoption-spurs-workforce-changes.aspx and 68% of professional services workers were using AI https://www.fastcompany.com/section/artificial-intelligence by February 2026. Thomson Reuters similarly found in spring 2026 that 74% of professionals https://www.thomsonreuters.com/en/institute/future-of-professionals-2026/report use AI several times a week for work. This widespread use shouldn’t surprise anyone. For accountants, attorneys, consultants, and other professionals, AI can accelerate research, summarize complex information, and assist with writing and other forms of communication. But its rapid adoption has a big downside—AI can produce information that sounds authoritative and looks polished while being completely wrong https://www.cnn.com/2023/08/29/tech/ai-chatbot-hallucinations . And when AI generates mistakes, responsibility for those errors belongs to the professional or business relying on the AI output—not companies that develop AI systems. AI errors have led to lawsuits, court fines, financial consequences and public embarrassment from fabricated citations, nonexistent sources, and inaccurate AI-generated information. These cases raise an increasingly important question: How can professionals benefit from AI without outsourcing the judgment their work requires? We are two certified public accountants https://www.xavier.edu/accounting-program/directory/kaitlin-newkirk who research tax accounting https://scholar.google.com/citations?hl=en&user=FJ9Y6QMAAAAJ . In the September 2026 issue of Journal of Accountancy, we examined what happened in Clinco v. Commissioner https://www.journalofaccountancy.com/issues/2026/sep/a-recipe-for-sanctions-ai-hallucinated-citations-in-tax/ , a case before the U.S. Tax Court where an attorney submitted a brief to the court with allegedly fake citations. The Tax Court is a specialized federal court that resolves disputes between taxpayers and the IRS. This case involved a restaurant owner accused of underreporting more than US$2 million in business receipts and claiming unsupported depreciation deductions on rental properties. The main issue concerned Clinco’s accounting records used to determine how much of their income to tax. The taxpayers’ attorney, Abraham Wagner, cited four cases to support his argument that an IRS notice assessing additional taxes was invalid. The court determined that three of the citations appeared to be AI-generated hallucinations that cited nonexistent Tax Court cases. The court rejected the taxpayer’s argument and warned Wagner that submitting fake citations could lead to sanctions from the court, including financial penalties or disciplinary action against Wagner. Ultimately, the court did not sanction Wagner in the case, but the decision stands to put attorneys and other professionals in Tax Court on notice that AI hallucinations will not be tolerated. AI errors made on behalf of clients are not limited to briefs drafted by attorneys. Nor are these mishaps limited to the United States. When Deloitte produced a 237-page report in 2025 https://www.theguardian.com/australia-news/2025/oct/06/deloitte-to-pay-money-back-to-albanese-government-after-using-ai-in-440000-report for the Australian government examining its automated system for penalizing welfare recipients who failed to meet job-seeking requirements, it contained nonexistent academic citations and a fabricated quotation it attributed to a federal court judgment. The global consulting firm later issued a revised version of the error-ridden report. Deloitte blamed generative AI for the errors in the original and agreed to partially refund its fees. Like most airlines, Air Canada has a chatbot on its website to respond to its customers’ questions. The chatbot told a customer in November 2022 that he could buy a plane ticket under a bereavement policy and be reimbursed later. This was not company policy, and Air Canada refused to honor what the chatbot had told the customer. The customer sued the airline https://www.bbc.com/travel/article/20240222-air-canada-chatbot-misinformation-what-travellers-should-know and won 812 Canadian dollars in damages. The British Columbia Civil Resolution Tribunal ruled that Air Canada had failed to take reasonable care to ensure that its AI bot provided accurate information and agreed with the customer that the airline had negligently misrepresented the bereavement policy. There have been other lawsuits filed against AI platforms and providers. In 2025, for example, a Georgia court dismissed a defamation lawsuit against OpenAI https://fingfx.thomsonreuters.com/gfx/legaldocs/egvblxokkvq/Walters%20v%20OpenAI%20-%20order.pdf after its ChatGPT tool falsely accused Mark Walters, a radio host, of embezzlement. The court ruled in favor of summary judgment for OpenAI and found Walters had not established fault and damages. There is also an ongoing case, Keene v. Google LLC https://www.siliconvalley.com/2026/09/16/google-ai-james-keene-defamation/ , that is similar to the Walters case. Keene, a TV producer, sued after Google’s AI summaries allegedly published false information about Keene’s criminal history. In September 2026, the court ruled that some of his defamation claims may proceed, without deciding yet whether Google was liable. Unfortunately, for anyone using AI to save time and use less labor, AI errors are often hard to spot. Unlike most obvious mistakes humans make, an AI-generated error may include supporting explanations, citations or references that make a passage appear complete and credible. The National Institute of Standards and Technology issued a report on AI risk management https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf in July 2024 that calls AI mistakes and hallucinations “confabulations”—the confident presentation of erroneous or false information. Some researchers draw a more specific distinction https://doi.org/10.1371/journal.pdig.0000388 between hallucinations, in which AI invents information outright, and confabulations, in which it fills gaps or distorts real information in a plausible way. Either kind of problem can be hard to detect without independent verification, but confabulations mimic actual research so closely they can trick AI users into believing the AI output is valid. A 2024 Stanford University study https://reglab.stanford.edu/publications/hallucination-free-assessing-the-reliability-of-leading-ai-legal-research-tools/ of specialized AI tools for legal research found erroneous results in 17% to 33% of the responses tested. Yet those findings and other grounds for concern about accuracy have not stopped many law firms from embracing AI tools due to their potential to save time and improve efficiency. Recent surveys indicate law firms and other employers have not adopted enough workplace safeguards to spot and root out AI-generated mistakes. A 2026 survey by the Nationwide insurance company https://news.nationwide.com/nationwide-survey-businesses-are-using-ai-faster-than-they-are-managing-its-risks/ found that only 36% of surveyed businesses have written AI policies, and only 37% provide any training on responsible AI use. As AI becomes more and more common in the workplace, we believe it’s clear that telling employees simply to “check the AI” is not enough. Businesses and the people working for them need clear expectations about when it’s appropriate to use AI, which outputs require independent verification, and who is ultimately responsible for the final work. AI may change how professional work gets done, but it does not change where the buck stops. Sarah Webber https://theconversation.com/profiles/sarah-webber-528143 is an associate professor of accounting at the University of Dayton https://theconversation.com/institutions/university-of-dayton-1726 . Kaitlin Newkirk https://theconversation.com/profiles/kaitlin-newkirk-2329817 is an assistant professor of accounting at Xavier University https://theconversation.com/institutions/xavier-university-4480 . This article is republished from The Conversation https://theconversation.com under a Creative Commons license. Read the original article https://theconversation.com/companies-may-be-held-liable-when-ai-harms-their-clients-and-customers-282234 .