A person falsely accused by a generative artificial intelligence system could seek economic damages for proven financial losses and non-economic damages for reputational harm, humiliation, and mental suffering. Presumed or punitive damages may also be possible, but usually require stronger proof of fault and depend heavily on state law. No leading U.S.
case has yet produced a damages award for this type of claim. On May 19, 2025, the Gwinnett County Superior Court granted OpenAI summary judgment after ChatGPT allegedly invented an embezzlement accusation about radio host Mark Walters, according to the court's order in Walters v. OpenAI.
Table of Contents
- Which economic losses could qualify?
- Which non-economic damages could be available?
- When are presumed or punitive damages possible?
- Why dissemination and fault can decide the case
- What evidence should a potential claimant preserve?
Which economic losses could qualify?
Economic damages compensate for measurable financial harm caused by the accusation. In defamation law, these losses are often called "special damages," meaning proven material or pecuniary losses rather than harm presumed from the statement itself. Possible examples include lost employment, canceled contracts, lost customers, reduced business revenue, or other expenses directly linked to the accusation. A claimant would need records showing both the amount and the causal connection.
Useful evidence may include: Reputational injury and emotional distress do not become special damages merely because a claimant assigns them a dollar amount. The U.S. Supreme Court distinguished pecuniary loss from those personal injuries in FAA v. Cooper.
- Termination notices or rejected job applications
- Canceled contracts and written explanations from clients
- Revenue records showing when losses began
- Messages from people who relied on the accusation
- Receipts for reasonable expenses incurred because of the incident
Which non-economic damages could be available?
Non-economic actual damages address real injuries that do not come with an invoice. They may cover impaired reputation and community standing, personal humiliation, mental anguish, and suffering. For example, a claimant might show that professional contacts stopped communicating, neighbors treated the person differently, or the accusation caused documented emotional distress. Witness testimony, contemporaneous messages, and treatment records may help establish those effects.
The claimant does not have to calculate every reputational or emotional injury with mathematical precision. However, competent evidence must show that the injury actually occurred. The Supreme Court explained both principles in Gertz v. Robert Welch.
When are presumed or punitive damages possible?
Presumed damages allow recovery without proof of a specific amount of actual injury. Punitive damages are intended to address especially blameworthy conduct rather than compensate for a measured loss. These categories face constitutional limits. Under Gertz, a private claimant who proves only a fault standard below knowledge of falsity or reckless disregard may recover actual injury, but not presumed or punitive damages.
A public official—and, under later doctrine, a public figure—faces a higher threshold for a defamatory falsehood involving protected subject matter. The claimant must prove "actual malice," which means knowledge that the statement was false or reckless disregard for whether it was true, as established in New York Times Co. v. Sullivan.
Why dissemination and fault can decide the case
A disturbing generated response does not automatically create a recoverable defamation claim. The claimant needs evidence that the accusation reached a third party, not merely that the system displayed false text to the person testing it. Fault also matters.
For a private individual, states may choose the governing fault standard, but the First Amendment does not allow defamation liability without fault. Public officials and public figures must satisfy the more demanding actual-malice rule where it applies. The Walters decision illustrates this practical barrier. A fabricated-looking output alone did not establish actionable defamation or damages, and the court entered summary judgment for OpenAI rather than awarding compensation.
What evidence should a potential claimant preserve?
Preserve the generated exchange in its original context. A cropped screenshot may omit wording that affects meaning, while an altered or incomplete record may invite disputes about authenticity.
Collect evidence in four categories: Save links, account details, timestamps, exports, correspondence, and unedited screenshots. Because state law can change the fault standard and available remedies, bring the preserved material to a qualified lawyer in the relevant state before records disappear or filing deadlines become an issue.
- The exact request, complete output, date, and system used
- Proof that another person received or viewed the accusation
- Evidence bearing on the defendant's fault
- Financial records and witnesses documenting resulting harm
You Might Also Like
- Severe Turbulence Causing Unrestrained Passenger Injuries: Which Economic and Non-Economic Damages Could Be Available?
- An AI Companion Allegedly Encouraging Self-Harm: Which Economic and Non-Economic Damages Could Be Available?
- A Wrongful Arrest Caused by Faulty Facial Recognition: Which Economic and Non-Economic Damages Could Be Available?