ChatGPT Sued for Libel: Is There Freedom of Speech for AI?

Do First Amendment protections apply to ChatGPT?

That is one of the key questions in a libel lawsuit filed in Georgia against ChatGPT-maker OpenAI by Mark Walters, radio host and founder of Armed American Radio. The gun rights advocate sued for defamation after the chatbot generated what he said were “false and malicious” statements about him.

The lawsuit marks the first time OpenAI would have had to defend ChatGPT against libel claims in court, according to Syracuse Law Review.

The case concerns a ChatGPT summary of a legal case that claimed Walters had been embezzling funds from the Second Amendment Foundation, a nonprofit supporting gun rights. Walters has no official relationship with the nonprofit. Journalist Fred Riehl, who asked ChatGPT for the summary, then informed Walters that the chatbot made up information about him.

Walters sued OpenAI, arguing that the generations were harmful to his career.

Recently, Gwinnett County Superior Court Judge Tracie Cason denied OpenAI’s motion to dismiss. In a one-page decision, the judge did not explain her decision, bar saying she had reviewed the applicable law.

The decision is not necessarily a harbinger of woe for OpenAI and generative AI tech in general, but does raise a complicated issue. Matthew D. Kohel, a partner at the law firm Saul Ewing, said the ruling “should not be relied on too heavily to predict the outcome of this case,” he said. “Motions to dismiss are predicated on a plaintiff’s allegations – rather than the facts produced by the parties to a lawsuit.”

Related:The New York Times sues OpenAI, Microsoft

Precedent-setting First Amendment case?

OpenAI is arguing that output from ChatGPT is not libel as the chatbot is not a publication. Moreover, Walters could not prove the system’s generation was created with malice, a key component for determining libel.

Every generative AI system hallucinates − generates a potentially false or misleading output. Notably, GPT-4, one of the underlying models powering ChatGPT, has one of the lowest hallucination rates of large language models, according to a Vectara study published last year. But it does still hallucinate.

The case could decide whether ChatGPT could be deemed a 'speaker' or 'publisher' with constitutional protections, or whether OpenAI is liable for false outputs generated by its own chatbot.

OpenAI’s tried to have the case thrown out for several reasons. The first being that there was “no publication.”

OpenAI said Riehl was the one who alerted Walters about the allegedly erroneous summary and it was he who “misused the software tool intentionally, and knew the information was false but spread it anyway,” which violated the startup's terms of service relating to spreading false information.

According to local statutes, Georgia law finds libel to occur when the communication is in print, writing, pictures or signs. Since the claims against Walters were not published, OpenAI denies any wrongdoing.

Absence of malice

OpenAI also argued that no malice took place. The landmark Supreme Court decision in New York Times v. Sullivan states that parties must prove libelous statements were made with "actual malice." Georgia state law defines libel as “a false and malicious defamation” that would harm a public figure’s reputation.

OpenAI said its AI-generated content is “probabilistic and not always factual,” adding that “there is near universal consensus that responsible use of AI includes fact-checking prompted outputs before using or sharing them.”

Kohel said Walters faces “an uphill battle” to prove actual malice. “ChatGPT is an AI tool and does not have knowledge or intent. And, because Walters’ claim is based on hallucinated output, he may have to show that someone at OpenAI input the erroneous information into ChatGPT’s training data with the necessary intent to defame him.”

OpenAI also took issue with Riehl’s use of the chatbot, contending that ChatGPT “repeatedly told Riehl that in this specific matter, it could not access, let alone accurately summarize, the legal document in question.”

Kohel added that that because Riehl knew the output was false “calls into the question whether Mark Walters suffered any reputational harm and damages.

The startup also highlighted its disclaimer in its terms of use, warning users that ChatGPT responses are not accurate and require verification.

However, John Monroe, attorney for Walters, told Ars Technica that "a disclaimer does not make an otherwise libelous statement non-libelous.”

OpenAI will now have to fight against a defamation suit brought by the “loudest voice in America fighting for gun rights.”

It already is facing other legal issues, including a copyright infringement suit brought by The New York Times. OpenAI rebuts the Times’ claims that it regurgitates its news stories nearly word-for-word and said it does not need to train on the Times' content per se. However, OpenAI did acknowledge that it cannot develop state-of-the-art models without access to copyrighted materials.

Adding to its legal woes is the FTC’s ongoing investigation into the startup. The agency is looking at how OpenAI obtained the data used to build its models.