One year after the decision in Walters v OpenAI (Case No. 23-A-04860-2) [pdf], the case remains one of the most significant judicial examinations of how traditional defamation principles apply to generative AI, albeit in a U.S. context.
While the claimant’s case failed on relatively conventional grounds, the judgment raises important questions about liability, publication, reasonable reliance, and the significance of AI disclaimers.
At a time when individuals and organisations are increasingly integrating generative AI into their everyday activities and business processes, the case is worth revisiting, not simply for what the court decided, but also for what it left unresolved.
The facts behind the claim
The dispute arose when journalist Frederick Riehl used ChatGPT to research a genuine firearms-related lawsuit, and asked ChatGPT to summarise the complaint involving the Second Amendment Foundation. After a series of unsuccessful attempts to obtain a summary of the case, ChatGPT generated a fictitious response falsely alleging that radio host and gun rights commentator Mark Walters had been involved in embezzlement.
The allegation was entirely fabricated.
Walters subsequently brought a defamation claim against OpenAI in the state of Georgia, arguing that the false output had damaged his reputation. In May 2025, however, the court granted summary judgment to OpenAI.
The court identified three independent reasons why the claim could not succeed. First, it concluded that a reasonable reader in the journalist’s position could not have understood the challenged ChatGPT output as communicating actual facts about Walters. In those circumstances, the output did not communicate a defamatory meaning as a matter of law. Secondly, he failed to establish fault under either the ordinary negligence standard or, as a public figure, the higher “actual malice” standard. Thirdly, he was unable to demonstrate any recoverable damages.
Although the decision was grounded in the specifics of Georgia defamation law and broader U.S. constitutional free speech principles, it remains one of the first and most significant judicial examinations of how traditional defamation principles may apply to AI-generated misinformation.
Why the court’s decision makes sense on the facts
Viewed narrowly (and leaving ethics/social considerations aside), the outcome in Walters is difficult to dispute under current U.S. law. The allegedly defamatory statement was communicated to a single recipient who was already aware of ChatGPT’s limitations, had encountered multiple warnings regarding potential inaccuracies during the research, and ultimately concluded that the output was false before acting on it. There was also little evidence of publication beyond that interaction and no evidence of tangible reputational harm.
On those facts, the decision arguably represents a straightforward application of established defamation principles rather than a radical reimagining of the law.
The “reasonable AI user”
A notable feature of the case was the court’s apparent focus on the perspective of a “reasonable user” of the AI system, which may impact how English Courts treat similar cases in future. Although the court did not establish a “reasonable AI user” test, significant weight was placed on the journalist’s familiarity with ChatGPT and his awareness of its known limitations.
The court concluded that a reasonable reader in those circumstances could not have understood the challenged output as conveying “actual facts”. The court’s reliance on repeated cautionary warnings and the user’s familiarity with ChatGPT suggests that clear disclosures regarding the fallibility of AI systems may play an important role in limiting potential liability. Whether courts will continue to attach the same significance to such disclaimers as generative AI becomes increasingly accurate, trusted and embedded in everyday and professional decision-making, however, remains an open question.
Would the same claim succeed in England and Wales?
The U.S. decision raises an obvious comparative question: how would a claim on similar facts be viewed in England and Wales?
Serious Harm
In Walters, the allegedly defamatory statement was communicated to a single individual who immediately recognised its inaccuracy, making it difficult to establish any meaningful reputational damage.
Similarly, in England and Wales, a claimant may struggle to satisfy the serious harm requirement under section 1 of the Defamation Act 2013 where a statement is communicated to only one person who does not believe it to be true. Demonstrating actual or likely reputational damage in such circumstances is likely to be challenging.
This issue may be particularly significant in the context of generative AI outputs, which are often dynamic and may vary substantially between users depending on the prompts entered, raising questions about the extent of publication, meaning and the degree of any resulting reputational harm. Where an allegedly defamatory output is seen by only a limited number of users, particularly users who recognise the information as potentially inaccurate, establishing serious harm may prove difficult. This however may give small comfort to individuals and businesses who worry about what may be said about them by generative AI.
Publication
AI outputs continue to challenge traditional concepts of publication and ultimate liability. English courts have previously grappled with similar issues in cases involving search engines, website operators and social media intermediaries, particularly when determining whether an intermediary, such as a social media platform, should be treated as a publisher of third-party content.
Determining who, if anyone, should properly be treated as the publisher of an AI-generated statement remains one of the most difficult questions in this area and one the court will no doubt have to consider in the near future.
In addition to identifying the publisher, the extent of publication presents a significant unresolved issue. Unlike traditional publications, AI-generated outputs are often generated in response to individual prompts and may be viewed only by a limited number of users (or even just be one individual user). Outputs can vary considerably between users, even where similar prompts are used, making it difficult to determine the scale of publication and the audience reached.
Meaning
The more trusted and popular AI becomes, the more likely it is that users will regard its outputs as accurate and authoritative, without independently verifying them or otherwise stress-testing the information received, despite any disclaimers that may be given. This shift in user expectations may affect the extent to which AI-generated statements are understood as conveying facts rather than speculation, error or merely generated text.
While the court in Walters concluded that a reasonable user in the circumstances could not have understood ChatGPT’s output as conveying actual facts, that assumption may become harder to sustain as generative AI becomes increasingly embedded in everyday and professional decision-making.
The more interesting question is not whether an English court would reach the same conclusion as the Georgia court, but whether it would do so for the same reasons.
Looking beyond Walters
One year on, Walters v OpenAI continues to represent an important early attempt to apply traditional defamation principles to generative AI. The decision suggests that existing legal frameworks may be capable of resolving at least some AI-related disputes without legislative intervention. Yet it also exposes limitations of applying established doctrines to a technology capable of generating convincing falsehoods (“hallucinations”), often without any intention that such statements be disseminated beyond the original interaction.
As AI systems become more sophisticated, trusted and deeply embedded in professional and personal decision-making, the real test for courts may not be cases like Walters, where the falsehood was quickly identified and caused little apparent harm. Rather, it may lie in claims involving greater reliance placed on AI-generated information, wider dissemination or republication (including through mainstream media and social media platforms), and more significant reputational consequences.
It may take time before clear judicial trends emerge across different jurisdictions. Whether established defamation law will prove sufficiently adaptable, or whether new legislative and regulatory solutions will ultimately be required, remains an open and increasingly important question.
Nataly Tedone is an Associate at Vardags specialising in global reputation, privacy and crisis management matters
Leave a Reply