“Inevitable Errors: Defamation by Hallucination,” Lyrissa Lydski

The rise of defamation-by-hallucination presents significant challenges for defamation law. In most states, the tort of defamation requires plaintiffs to establish that the defendant was at least negligent in publishing a defamatory, false, factual statement about them; depending on the state, and sometimes on the circumstances of publication, plaintiffs may also have to prove some degree of reputational damage to recover compensation. As a matter of constitutional law, plaintiffs who are public officials or public figures must prove that the defendant knew before publishing that the defamatory statement was false or recklessly disregarded its truth or falsity. 

To state the obvious, it is not possible to judge what a Large Language Model, or LLM, “knew” prior to publishing an output in response to a prompt. Nor is it possible to determine whether the LLM exercised reasonable care, that is, the standard of care a reasonable person would exercise under the same or similar circumstances.

State courts have long adapted tort doctrines to achieve what they perceive to be sound communications policy. This was true even before the Supreme Court intervened to balance state interests in protecting reputation against free speech concerns. Adapting defamation law to generative AI is simply the newest challenge in defamation law’s long and ongoing evolution.   

Nonetheless, the stakes are high. If defamation law imposes liability on AI companies for every defamatory hallucination, innovation may suffer. The costs of defending against numerous claims and the threat of massive, unpredictable damage awards may cause developers to adopt more conservative designs, limit AI capabilities, and implement excessive filtering and content moderation. “Getting it wrong” could hamper responsible AI development, shifting costs to end users and reducing accessibility. In addressing defamation by hallucination, the law must balance the desire to impose accountability for reputational harm with the need to foster the development of a promising medium of communication.

Scholars have begun to offer tentative solutions. Notably, Eugene Volokh has suggested that AI companies should not be held liable for defamation until they are put on notice of false and defamatory outputs and fail to correct them. Volokh’s proposal, however, may lead to undesirable gamesmanship and excessive content removal, even if it is technologically feasible. 

This article proffers an alternate approach. First, the article demonstrates why AI output is valuable from a First Amendment perspective by both summarizing the most compelling arguments made by other scholars as well as focusing on one that has been neglected—namely, the distinctive First Amendment value of AI as an information-gathering tool. Having explained the weight AI outputs should be given in the balance against reputation as a general matter, the article then applies existing defamation doctrines—both common law and constitutional—to AI speech. This analysis highlights the points at which existing doctrine might be tweaked to accommodate communications policy concerns. The most likely contenders for adaptation are the elements of publication, fault, damages, and perhaps privileges. After explaining why notice and takedown based liability for defamatory hallucinations inadequately protects reputation, the article offers a “tweaked” prima facie case to properly balance innovation with accountability and reputation with free expression.

Lyrissa Lidsky is the Raymond & Miriam Ehrlich Chair in U.S. Constitutional Law at the University of Florida’s Levin College of Law. Her research focuses on the intersection of the First Amendment and Tort Law, with an emphasis on free speech and new technologies. Lidsky is co-reporter for the Third Restatement of Defamation and Privacy and is the new author of "Sack on Defamation." She has co-authored a leading Media Law casebook, and has published additional books and dozens of law review articles. Her most recent article, First Amendment Disequilibrium, appeared in the Virginia Law Review. Her work has been cited by many U.S. courts and the highest courts of Canada and Hong Kong. She was Dean of the University of Missouri Law School from 2017-2022. She graduated from the University of Texas Law School with high honors and clerked on the US Court of Appeals for the Ninth Circuit.