Courts Put AI Errors on Companies, Not AI Developers

By Daniel Okafor

A U.S. Tax Court attorney escaped sanctions in 2026 after three of the four cases he cited turned out not to exist. That ruling in Clinco v. Commissioner signals who pays when AI gets it wrong: the professional who relied on it, not the company that built the tool.

The case is one of a growing pile. Courts, regulators and clients are assigning responsibility for AI errors to businesses and professionals, according to an analysis by Sarah Webber, associate professor of accounting at the University of Dayton, and Kaitlin Newkirk, assistant professor of accounting at Xavier University. The two certified public accountants examined the Tax Court dispute in the September 2026 issue of Journal of Accountancy.

How common is workplace AI?

Very. Gallup found that 52% of U.S. employees and 68% of professional services workers were using AI by February 2026. Thomson Reuters separately found in spring 2026 that 74% of professionals use AI several times a week for work.

The appeal is obvious. AI accelerates research, summarizes dense material and drafts communication for accountants, attorneys and consultants. The catch: it produces output that sounds authoritative and looks polished while being completely wrong.

What happened in the Tax Court case?

The dispute involved a restaurant owner accused of underreporting more than US$2 million in business receipts and claiming unsupported depreciation deductions on rental properties. The core issue concerned the accounting records used to determine taxable income.

The taxpayers' attorney, Abraham Wagner, cited four cases to argue that an IRS notice assessing additional taxes was invalid. The court determined that three of the citations appeared to be AI-generated hallucinations citing nonexistent Tax Court cases.

The court rejected the argument and warned Wagner that submitting fake citations could lead to sanctions, including financial penalties or disciplinary action. It ultimately did not sanction him. But the decision puts attorneys and other professionals on notice that AI hallucinations will not be tolerated.

What do AI failures cost companies?

The damage is not confined to courtrooms or to the United States.

• Deloitte, Australia, 2025. The firm produced a 237-page report for the Australian government on its automated system for penalizing welfare recipients. The report contained nonexistent academic citations and a fabricated quotation attributed to a federal court judgment. Deloitte blamed generative AI, issued a revised version and agreed to partially refund its fees.
• Air Canada, 2022. The airline's chatbot told a customer in November 2022 that he could buy a ticket under a bereavement policy and be reimbursed later. That was not company policy, and the airline refused to honor it. The customer sued and won 812 Canadian dollars in damages. The British Columbia Civil Resolution Tribunal ruled Air Canada failed to take reasonable care to ensure its bot provided accurate information and had negligently misrepresented the policy.

When can AI developers be sued?

Sometimes, but the record favors the platforms. In 2025, a Georgia court dismissed a defamation lawsuit against OpenAI after ChatGPT falsely accused radio host Mark Walters of embezzlement. The court granted OpenAI summary judgment, finding Walters had not established fault and damages.

Keene v. Google LLC is still moving. TV producer Keene sued after Google's AI summaries allegedly published false information about his criminal history. In September 2026, the court ruled that some of his defamation claims may proceed, without deciding whether Google is liable.

Why are AI errors so hard to catch?

A 2024 Stanford University study of specialized AI tools for legal research found erroneous results in 17% to 33% of responses tested. Yet law firms keep adopting the tools for their time savings.

The National Institute of Standards and Technology issued a report on AI risk management in July 2024 that calls AI mistakes "confabulations" — the confident presentation of erroneous or false information. Some researchers draw a sharper line: hallucinations invent information outright, while confabulations fill gaps or distort real information in a plausible way. Confabulations mimic actual research so closely that they can trick users into believing the output is valid without independent verification.

Are businesses prepared?

Mostly not. A 2026 survey by the Nationwide insurance company found that only 36% of surveyed businesses have written AI policies, and only 37% provide any training on responsible AI use.

Webber and Newkirk argue that telling employees simply to "check the AI" is not enough. Businesses need clear expectations about when it is appropriate to use AI, which outputs require independent verification, and who is ultimately responsible for the final work. As adoption climbs toward saturation, the companies that answer those questions first will face the fewest of these courtrooms.