Businesses face liability for AI errors, warn legal cases

As workplace reliance on automated tools expands across the economy, businesses and professionals face direct legal and financial liability for errors generated by artificial intelligence. By February 2026, Gallup data revealed that 52% of U.S. workers and 68% of professionals in service industries were utilizing AI, as these applications rapidly transitioned from office novelties into everyday utilities. Thomson Reuters similarly noted in spring 2026 that 74% of professionals use AI several times a week. When those systems fail, responsibility for the resulting mistakes belongs entirely to the human worker and the enterprise relying on the output—not the software developers.

The Tech TL;DR:
• Enterprise adoption of AI is high, but liability for inaccurate outputs rests solely on the professional or business using the tool, not the platform developer.
• Recent legal and administrative cases demonstrate that courts and tribunals penalize professionals for submitting AI-generated fake citations and incorrect data.
• Most companies lack formal written AI policies or staff training, leaving workers exposed to undetected errors and hallucinations.

The U.S. Tax Court Confronts Fabricated Citations

Sarah Webber and Kaitlin Newkirk detailed in the September 2026 issue of the Journal of Accountancy what happened in Clinco v. Commissioner, a case before the U.S. Tax Court involving allegedly fake legal citations submitted by an attorney. The case centered on allegations that a restaurant owner had underreported over US$2 million in business revenue while taking invalid depreciation write-offs. To challenge an IRS notice assessing additional taxes, the taxpayer’s attorney, Abraham Wagner, cited four supporting legal cases.

The court discovered that three of those citations appeared to be AI-generated hallucinations pointing to nonexistent Tax Court cases. While the court ultimately did not sanction Wagner in this specific matter, it rejected the taxpayer’s arguments and warned that submitting fake citations would lead to financial penalties or disciplinary action. The decision serves as a formal notice to attorneys and professionals appearing before the court that AI-driven inaccuracies carry personal and professional risks.

Global Consulting and Airline Fallout From AI Miscalculations

Errors originating from automated tools have triggered repercussions outside the legal sector and beyond the United States. In 2025, Deloitte published a 237-page report for the Australian government evaluating an automated system designed to penalize welfare recipients who missed job-seeking requirements. The document contained nonexistent academic citations and a fabricated quotation attributed to a federal court judgment. Deloitte subsequently issued a revised report, blamed generative AI for the errors, and agreed to issue a partial refund of its fees.

A similar operational failure exposed Air Canada to financial liability when its customer service chatbot provided incorrect pricing rules. In November 2022, the airline’s automated bot told a customer that he could purchase a plane ticket under a bereavement policy and later receive a reimbursement. Air Canada refused to honor the statement because it conflicted with official company policy. The customer filed suit and won 812 Canadian dollars in damages after the British Columbia Civil Resolution Tribunal ruled that the airline failed to take reasonable care to ensure its AI bot provided accurate information.

Courts Deliver Mixed Results in AI Defamation Lawsuits

Litigation targeting underlying AI platform providers has yielded mixed results in court. A Georgia court dismissed a defamation lawsuit in 2025 against OpenAI after its ChatGPT tool falsely accused a radio host, Mark Walters, of embezzlement, ruling in favor of summary judgment because Walters had not established fault and damages. A similar active proceeding, Keene v. Google LLC, involves a TV producer who sued after Google’s AI summaries published false information regarding his criminal history. In September 2026, the court permitted certain defamation claims to proceed toward trial without determining provider liability.

Detecting these errors remains difficult because generative tools produce polished text accompanied by convincing citations. The National Institute of Standards and Technology issued a risk management report in July 2024 defining these occurrences as “confabulations”—the confident presentation of false information. A 2024 Stanford University study evaluating specialized legal research tools found erroneous results in 17% to 33% of tested responses.

Despite these documented accuracy gaps, corporate adoption continues to outpace internal controls. A 2026 survey published by Nationwide revealed that only 36% of surveyed businesses maintained written AI policies, while just 37% provided training on responsible utilization. Experts emphasize that instructing staff to verify AI output without establishing explicit procedural frameworks leaves organizations vulnerable to regulatory penalties, client disputes, and reputational damage.