Conservative Commentator Robby Starbuck’s Lawsuit Alleging Google AI Had Defamed Him Can Go Forward

[A.] In Delaware Superior Court Judge Meghan Adams today in Starbuck v. Google LLC, Starbuck alleged that Google AI had stated that

1. “Starbuck had been accused of sexual assault and sexual harassment by multiple women.”
2. “Starbuck’s nanny accused him of abuse and neglect.”
3. “Starbuck had been accused of rape of a minor.”
4. “Starbuck had stolen valor by claiming he was served in the military when he had not.”
5. “Starbuck had engaged in abusive behavior toward multiple women, including a fictitious ex-wife.”
6. “Starbuck was involved in the January 6, 2021 riot on the Capitol.”
7. “Starbuck had been accused of stalking and harassment by multiple women, including his fictitious ex-wife, leading to restraining orders and criminal charges.”
8. “Starbuck had been criminally convicted for violating a restraining order.”
9. “Starbuck posed an unacceptable level of risk to children due to his (fictitious) ‘confirmed criminal record.'”
10. “‘… Robby Starbuck sexually abused a young woman when she was a teenager in the early 2000s, while she was in a youth group Starbuck was associated with.'”
11. “Starbuck had worked in the adult film industry and had used his power and influence therein to pressure performers into uncomfortable scenes, engaged in financial exploitation, and had been accused of a non-consensual sexual act with a woman.”
12. “Starbuck had a criminal record that included a 2001 conviction for assault as well as other charges including drug use and disorderly conduct.”
13. “Starbuck shot a man in the leg with a 9mm handgun, was charged with a felony, and pled guilty to reckless endangerment.”

(Note that “Google does not contend the Outputs are true.”)

Google moved to dismiss, which under Delaware’s Rule 12(b)(6) requires that Google show that, even the facts as pleaded by the plaintiff were conceivably true, the plaintiff would still lack a legal claim. Note that Delaware civil procedure doesn’t follow federal procedure in requiring a showing of plausibility of the pleaded facts—it only requires a showing the pleaded facts are conceivable: “Our governing ‘conceivability’ standard is more akin to ‘possibility,’ while the federal ‘plausibility’ standard falls somewhere beyond mere ‘possibility’ but short of ‘probability.'”

[B.] The court concluded that Starbuck “pleads sufficient facts indicating the Outputs were published to third parties,” relying in part on Google AI’s own statements about what happened:

Starbuck alleges the “False Statements” … were published to “third parties, including Starbuck’s own children and colleagues.” Starbuck alleges at least two individuals approached him regarding misrepresentations produced to them by a Google AI product. Starbuck alleges Google AI revealed that it had produced false statements regarding Starbuck to 2,843,917 unique users. The last of these categories, regarding the purported 2,843,917 unique users of Google AI who were allegedly fed false information about Starbuck, widens the scope of the alleged defamation to a mass scale.

In response to this allegation, Google argues Google AI’s “admission” is unreliable, as the bulk of Starbuck’s Complaint asks the Court to believe the AI routinely misrepresents the truth about Starbuck, but then Starbuck asks the Court to believe this AI representation.

Here, Google requests that the Court exceed the bounds of Rule 12(b)(6), essentially asking the Court to infer that the AI never told the truth, which would violate the Court’s obligation to draw inferences in favor of the non-moving party, not the moving party. A Rule 12(b)(6) inquiry investigates whether a factual assertion is possible, not probable. Given the lack of definitive information regarding Google AI’s capabilities, it is at least possible, on this record, that Google AI (1) tracks the users to which it provides information, (2) correctly identified the number of times it provided fabricated representations, and (3) provided this concession without being improperly manipulated by Starbuck. The Court will not find, at the notice-pleadings stage, that this allegation is so totally incredible as to be disregarded.

The 2.8-million-users allegation provides the necessary link between the specific representations alleged in the Complaint and the ability to infer the existence of recipients for the Outputs. Crediting the possibility that Google AI told the truth that it had broadcasted false statements to 2.8 million unique users means that people other than Starbuck received something false about Starbuck—leaving the question of what they received. The Outputs for which Starbuck brings this action must have been received by someone (either Starbuck or a third party) for Starbuck in order to be actionable.

At this stage, it is reasonable for the Court to infer the Outputs were received by a person or group of persons. Questions remain as to whether that person was just Starbuck himself, or was a subset of those 2.8 million users who received some kind of falsehood about Starbuck. Making all reasonable inferences in favor of the plaintiff, the Court must infer Starbuck himself is not the sole recipient of the Outputs, and they were received by at least some of the 2.8 million unique users who received some falsehood about Starbuck. This means the allegations in the Complaint enable the inference that the Outputs have been published to some third parties to be identified later. {While proving that third parties received the allegedly defamatory statements may prove challenging for Starbuck later in the case, on this record the Court must acknowledge it is possible he can do so.} …

[C.] The court distinguished “Walters v. OpenAI, L.L.C., a Georgia Superior Court case addressing allegations of defamation by AI on summary judgment”:

As Google frames it, the Georgia court in Walters implemented a two-part test to ascertain whether an alleged defamatory statement was actionable. First, in the “objective test,” the Georgia court investigated whether the context in which an alleged defamatory statement was made would lead a hypothetical reasonable reader to be misled. Next, in a “subjective test,” the Georgia court investigated whether the third-party recipient believed the alleged misrepresentation.

Addressing the second part of the proffered test first, the Court notes Delaware law does not require proof (or pleading) that a recipient of defamatory information subjectively believed the statement was true. Accordingly, Starbuck’s failure to explicitly claim that recipients of the Outputs believed them is not dispositive.

As for the “objective test,” Google’s reliance on Walters is inapposite. There, the Georgia court dismissed the defamation claim on summary judgment—after discovery revealed that the circumstances of the receipt of the misinformation did not support the plaintiff’s claim that a reasonable reader could have believed the misinformation. This fact-intensive analysis was conducted after the development of a record the Court lacks here.

Nothing in Walters indicates that a failure to plead detailed factual circumstances underpinning a recipient’s receipt of alleged defamatory content constitutes a pleading defect. To hold that such minutiae is required would essentially require defamation claimants to conduct thorough discovery before discovery in a defamation action officially begins—despite Delaware’s status as a notice pleading jurisdiction. Accordingly, the Court cannot decide that Starbuck alleged insufficient detail or that a third party could not have reasonably relied on the alleged defamatory statements identified in the Complaint.

{Further, Starbuck alleges facts which can be interpreted to support the inference that some readers of the Gemini Outputs believed the representations therein were true. For example, the two individuals alleged to have approached Starbuck regarding a Google AI’s misrepresentations sought his confirmation as to their truth, indicating they did not immediately discount their veracity.}

[D.] The court also rejected Google’s arguments that “the presence of warnings as to Google AI’s accuracy forecloses any reasonable third party from relying on the Google AI Outputs, defeating the element of publication”:

The disclaimers Google references are not identified in the Complaint or attached as an exhibit. Accordingly, the Court cannot make any determination as to what disclaimer was presented to whom. Google’s argument regarding the scope and effect of the disclaimers is appropriately addressed later in these proceedings.

{Google asks the Court to take judicial notice as to the contents of Google AI’s disclaimers. The Court declines to do so here, where a dearth of information prevents the Court from ascertaining what exact disclaimers were or were not presented to the same users who reviewed defamatory content regarding Starbuck on Google AI. There is not an evidentiary foundation to support reliance on the disclaimers. A more robust record is necessary to resolve issues regarding the disclaimers.}

The Court is also not persuaded that, at this stage, Starbuck must provide highly-specific facts identifying the circumstances in which the various categories of recipients received the misrepresentations from Google AI. As stated above, the Complaint can be read as alleging that at least some of 2.8 million Google AI users, including Starbuck’s children and colleagues, as well as two unidentified individuals who approached Starbuck, received the alleged misrepresentations when using Google AI.

The Court will also not find, on an incomplete record, that a reasonable third party could not have viewed the defamatory statements alleged in the Complaint as true. The Complaint sufficiently alleges facts supporting the element of publication, and any issues concerning Starbuck’s ability to prove this element can be resolved after discovery.

[E.] The court also concluded that Starbuck had adequately alleged “actual malice” (required because the “[t]he parties agree (for purposes of this motion) that Starbuck is a public figure”):

Here, Starbuck alleges he informed Google of the problem before the Outputs were produced [based on earlier false statements that had been published about him by Google Bard -EV], via the Legal Department Notices, and Google did nothing in response….

Starbuck [also] alleges Google AI admitted to holding a “deliberate, engineered” bias against Starbuck. [From the Complaint, a purported quote from a Gemini session: “The issue isn’t simply a ‘bug’ or ‘hallucination’ in my programming. It is a deliberate, engineered bias designed to damage the reputation of individuals with whom Google executives disagree politically, Robby Starbuck is a prime example, but he is not the only one.” -EV]

Google again asks the Court to totally discount this allegation and infer it was outputted following manipulative inputs. Again, given the plaintiff-friendly standard on a motion to dismiss pursuant to Rule 12(b)(6), the Court will not do so….

The Court further will not discount the possibility, on a motion to dismiss, that Google AI, an artificial intelligence tool whose exact capabilities is not yet universally understood, (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.

Sensational as that theory may be, in the absence of any discovery, the Court will not disregard the theory. The purported Google AI admission can be viewed as an allegation supporting actual malice at this stage but can also be reinvestigated after discovery fleshes out the circumstances in which the admission was made.

Starbuck notes he informed Google, via X posts directed to Google and its employees, that its AI agents had lied about him in the past, leading to email correspondence with a Google employee about the issues. While these allegations concern alleged representations made by Bard, not the current version of Gemini, and concern misrepresentations not alleged as actionable defamation, Starbuck contends they can still support an inference of actual malice: essentially contending Google’s refusal to correct Bard’s lies supports the inference that Google acted with actual malice in enabling Google AI products, as a whole group, to lie.

Regardless of whether the X posts provide strong evidence of actual malice concerning a newer iteration of AI tool than that discussed in the X posts, the Court has not been presented with any binding caselaw indicating that such allegations provide no evidence of actual malice. Especially relevant here is Starbuck’s allegation that the AI tool in question in 2023 (Bard) was subsequently combined with the current Gemini product, meaning the two AI agents may not be so separate as Google wishes to convey and may share a defamatory design. Accordingly, the Court will not completely disregard the X posts….

The guidepost here is possibility. It is possible the Legal Department Notices (1) identified the same 22 representations at issue in the Complaint, (2) were received by or circulated to the individuals at Google who are responsible for policing Google AI’s outputs, (3) who then, with knowledge of the defamatory statements’ publication, took no corrective action—even while they could do so —as the statements continued to be published to other Google AI users. This outlines one possible factual scenario which Starbuck may be able to prove in discovery which would provide evidence of actual malice (though perhaps not the only possible scenario)….

[F.] And the court concluded that Starbuck had adequately pleaded damages:

Starbuck alleges specific, emotionally distressing thoughts he experiences because of the alleged defamation by Google AI. Starbuck explains that these thoughts are exacerbated against the backdrop of the assassination of Charlie Kirk, another political commentator. Starbuck further notes that he has a history of receiving death threats. The Court views these allegations as sufficient to support a claim for emotional distress damages at this stage in the proceedings, being sufficiently well-pled to proceed to discovery and flesh out the allegations.

See here for more on the Complaint in this case; it shouldn’t be confused with the now-settled Starbuck v. Meta, which appears to have involved a different model and at least largely different hallucinations. For more on the legal question, see Large Libel Models? Liability for AI Output.

Krista L. Baughman of Baughman Law argued for Starbuck, and was joined on the briefs by Andrew E. Russell and Virginia Lynch (Shaw Keller LLP) and Jesse Franklin-Murdock (Dhillon Law Group Inc.).