Delaware judge lets landmark AI defamation suit against Google proceed

Google Must Face AI Defamation Suit. Discovery Into Model Design Is on the Table.

When a chatbot fabricates a claim that you committed sexual assault, links you to a white supremacist, and suggests arguments for your execution, who is legally responsible? On July 24, a Delaware judge gave an answer that should make every company deploying a large language model reconsider its exposure.

Judge Meghan A. Adams of the Delaware Superior Court ruled that Google must face a defamation lawsuit brought by conservative activist Robby Starbuck, who alleges that Google's Bard AI chatbot produced damaging falsehoods about him, including fabricated sexual assault accusations, an incorrect connection to white supremacist Richard Spencer, and arguments in favor of his execution 1.

Bard, which Google has since replaced with its Gemini model, allegedly produced what Starbuck's lawsuit called "radioactive lies," portraying him as a sex offender and campaign finance criminal 1. Starbuck, an anti-DEI crusader who has organized social media campaigns against companies including Walmart and Harley-Davidson, filed the suit in October 2025 1.

The case survived a motion to dismiss. That is the detail that matters most. When a defamation claim clears that threshold, it enters discovery, the phase where plaintiffs can demand internal documents, training data, and testimony about how a model was constructed. For any company shipping an LLM-powered assistant, discovery is where the real legal risk lives.

Judge Adams acknowledged that the dispute opened a new frontier for defamation law, one where AI tools are "allegedly employed to effectuate the defamatory ends of their makers" 1. But she framed the immediate question in ordinary terms: at the motion-to-dismiss stage, standard defamation precedent governs, not novel questions about machine autonomy.

The sleeper allegation in the ruling is not the defamation claim itself. It is Starbuck's charge that Google deliberately engineered bias against him into the model, an assertion he says he derived from Bard's own output 1. Judge Adams declined to throw it out, writing that it would be "premature to discount the possibility" that Google's AI can "review and understand its own programming" 1.

"Sensational as that theory may be," she added, "in the absence of any discovery, the court will not disregard the theory" 1.

That paragraph could reshape how courts treat AI training. If the bias claim survives summary judgment, discovery could force Google to produce internal communications about how Bard was trained, what guardrails were in place, and whether human reviewers made deliberate editorial choices about what the model would say about public figures. For any company that tunes an LLM, the prospect of a court examining training data and alignment decisions is a legal exposure that no terms-of-service clause can prevent.

Delaware, home to roughly two-thirds of Fortune 500 companies, has become a preferred venue for high-profile defamation litigation 1. The same court recently allowed California Governor Gavin Newsom's $787 million defamation suit against Fox News to proceed 1. Judges in this venue are comfortable with complex defamation claims, and the Starbuck ruling now extends that willingness to cases where the speaker is a language model rather than a human.

The practical implications for enterprises are direct. AI defamation liability is no longer hypothetical. A chatbot that fabricates accusations about a real person can expose its operator to the same legal process as a traditional publisher, and the discovery that follows can reach into the model's design. Companies deploying generative AI assistants should treat output accuracy as a legal risk, not just a product quality issue. Red-teaming for defamatory outputs about identifiable individuals, maintaining audit logs of model behavior, and understanding that a motion to dismiss is not a dependable shield are now baseline operational requirements, not optional safeguards.

Starbuck previously filed a similar case over a Meta Platforms chatbot that falsely connected him to the January 6 attack on the US Capitol, but he dropped that case 1. The Google case, numbered N25-C-10-211 in Delaware Superior Court, proceeds to discovery 1.

The ruling does not settle whether AI-generated falsehoods count as defamation the way human speech does. It establishes that the question will be litigated rather than dismissed. For companies betting that a chatbot's hallucinations are someone else's legal problem, that distinction is the one that will determine whether generative AI deployment remains a calculated risk or becomes a liability they cannot insure away.