Robby Starbuck has been dubbed a “conservative influencer” and “one of the flashier lights in a constellation of anti-DEI media personalities.” He is “known for targeting corporate DEI programs,” with businesses such as Tractor Supply, John Deere, and Harley–Davidson caught in his campaigns’ crosshairs.
However, Starbuck ultimately may become better known—at least in media law circles—for something very different: testing the boundaries of libel law when artificial intelligence tools allegedly produce false, reputation-harming statements about public figures. Starbuck’s defamation lawsuit against Google recently survived the company’s motion to dismiss (MTD) in Delaware state court.
As Judge Meghan Adams encapsulated it in her July opinion, Starbuck’s case centers on more than 20 outputs allegedly generated by various “Google AI” tools such as Gemini. The responses link Starbuck to criminal and sordid activities, including outputs stating that he has “been accused of sexual assault and sexual harassment by multiple women,” took part “in the January 6, 2021 riot on the Capitol,” was “criminally convicted for violating a restraining order,” and “shot a man in the leg with a 9mm handgun, was charged with a felony, and pled guilty to reckless endangerment.” Starbuck contends these outputs are “provably false, being fabricated from whole cloth.”
As a public-figure plaintiff, Starbuck must prove that Google acted with actual malice in generating false statements about him. Actual malice in defamation law doesn’t mean publishing a statement with ill will or hatred toward a plaintiff. Rather, it focuses on a defendant’s state of mind about the false statements at the time they were published: Did the defendant either know the statements were false or did the defendant act with reckless disregard for whether they were false? Reckless disregard means a defendant had a high degree of awareness the statements were probably false—entertained serious doubts about their veracity—but published them anyway.
Starbuck’s complaint features what Adams aptly calls a “sensational” theory to prove that Google acted with actual malice. He contends a Google AI tool, apparently in response to a prompt, “admitted to having a ‘deliberate, engineered bias designed to damage the reputation of individuals with whom Google executives disagree politically,’ including Starbuck.” In non-legalese, Starbuck alleges that an AI tool ratted out its maker on whether output about Starbuck was published with reckless disregard for its veracity.
For this purported-admission logic to hold up after discovery (interrogatories, depositions, and document production) concludes, Adams indicated that Starbuck must prove that the Google AI tool that generated the alleged admission and falsities
(1) can review and understand its own programming and design, (2) therefore knows it has been “deliberately engineered” to defame individuals such as Starbuck, and (3) accurately reported that to the individual who prompted it to do so as alleged in the Complaint. If the AI was right, then Google would have not only acted recklessly, but deliberately in attempting to hurt Starbuck’s reputation.[CC1]
Google maintains Starbuck engaged in “a deliberate effort to induce purportedly defamatory hallucinations” from its AI tools. It emphasizes that he “failed to provide the Court with any of the inputs that were used to generate these purported ‘admissions’ or the services he accessed.” Google adds that “all AI systems are capable of error, and the specific services at issue here were misused and induced to hallucinate.”
We’re now six decades and vast factual miles removed from the underpinnings of New York Times Co. v. Sullivan, the opinion in which the US Supreme Court adopted the actual malice standard. The question there was whether a print newspaper’s personnel knew about (or acted recklessly) regarding falsities in a full-page advertisement that a Montgomery, Alabama, commissioner claimed defamed him. Starbuck ultimately must prove, as Adams wrote, “that the persons responsible at Google” for “policing” a chatbot’s output acted with actual malice.
It’s still early in the case—discovery hadn’t occurred when Adams ruled—and surviving an MTD in Delaware is rather easy. As Adams wrote, it’s a “plaintiff-friendly standard” that merely requires a claim to be “reasonably conceivable” and “possible, not probable.” (Emphasis in original.) Furthermore, judges must make “all reasonable inferences in favor of the plaintiff.”
Surviving a summary judgment motion filed by Google—after Starbuck is deposed, discovery concludes, and facts are revealed regarding how exactly the outputs at issue were generated—will probably prove more difficult for Starbuck. Walters v. OpenAI, another case targeting allegedly defamatory statements produced by a chatbot, didn’t clear summary judgment, as I’ve explained. Multiple warnings and disclaimers provided by OpenAI alerting users that ChatGPT can produce false statements helped OpenAI defeat that case, and Google’s warnings about its AI tools may ultimately do the same in Starbuck. For now, however, a “sensational” theory for proving actual malice remains in play, with Starbuck filing an amended complaint adding more causes of action.