A Delaware Superior Court judge refused to dismiss a defamation lawsuit against Google over content generated by its Bard AI chatbot, and the three arguments Google made in its defense are the same arguments most enterprise AI vendors would make if your tool started generating false statements about people.
On July 24, 2026, Judge Meghan A. Adams denied Google's motion to dismiss in Starbuck v. Google LLC (No. N25C-10-211). Conservative activist Robby Starbuck filed the suit in October 2025, alleging Bard generated statements calling him a "child rapist," "serial sexual abuser," and "shooter," and linked him to white nationalist figure Richard Spencer. Starbuck is seeking over $15 million in damages. The case now proceeds to discovery.
TL;DR: On July 24, 2026, Delaware Superior Court denied Google's motion to dismiss in Starbuck v. Google (N25C-10-211). The court rejected Google's three main defenses against an AI defamation claim -- that AI output is not "publication," that no specific viewer could be identified, and that experimental disclaimers provide cover. Enterprise teams deploying AI tools that generate content about named individuals need to update three things: complaint response procedures, content review policies, and their legal assumptions about Section 230.
Judge Adams described the case as opening "a new frontier for defamation law, in which artificial intelligence tools are allegedly employed to effectuate the defamatory ends of their makers." That is careful language from a court that is not ready to say Google is liable, but is ready to say the claim is plausible enough to investigate. The distinction matters for every team running an AI tool that could generate a sentence about a named person.
What Google Bard allegedly said
The factual record matters here because the specific nature of the Bard outputs shapes the legal theory. According to the complaint, Bard generated statements calling Starbuck a "child rapist," linking him to Richard Spencer, and suggesting arguments "in favor of his execution." These are not minor inaccuracies or factual disputes. They are statements that, if about a real person and false, would qualify as defamation per se under most jurisdictions' standards -- categories of false statement so obviously harmful that the law presumes damage without requiring proof.
Starbuck's complaint alleges the false outputs began circulating as early as 2023 and continued across later iterations of Google's AI models. He says he became aware of the defamatory statements while using Bard in 2023. The complaint describes a pattern of outputs that Google's systems continued to generate even after the alleged defamatory nature of those outputs was brought to the company's attention.
This is the fact the court found most significant: what happened after Starbuck raised the problem, not just that the problem existed. The case is not about whether AI sometimes generates false information. Every enterprise team already knows it does. The case is about what legal consequences attach when a company continues to serve that false information after someone tells them it is false and harmful.
The three defenses Google lost
Google's motion to dismiss raised three arguments that will sound familiar to anyone who has thought about AI liability:
Argument 1: AI output is not "publication." Google argued that because users generate AI responses through their own queries, the company is not "publishing" the outputs -- users are. The court was not persuaded. The standard for publication in defamation is whether a defendant communicated false information to at least one third party. A user who searches and receives a Bard response is that third party. The fact that a query triggers the output does not move the publication act from Google to the user.
Argument 2: No specific viewer can be identified. Google argued that Starbuck could not prove any specific person saw and relied on the defamatory outputs. Delaware defamation law does not require the plaintiff to name every viewer of a defamatory statement. Publication to any third party is sufficient. The court found that the allegation that users received Bard outputs -- even without naming each one -- satisfied the publication element at the motion to dismiss stage.
Argument 3: Experimental disclaimers shield the company. Google pointed to Bard's disclaimers that outputs may be inaccurate and should not be relied upon. The court found that disclaimers are relevant to later stages of the analysis -- particularly questions of reliance -- but do not by themselves negate the defamatory nature of an output. A company cannot generate a statement calling a specific named person a child rapist and then disclaim its way out of the legal consequences.
None of these arguments worked at the motion to dismiss stage. That does not mean Google will lose the case. Discovery may reveal facts that support Google's defenses on the merits. But the court found the claims were legally viable, and that is the signal enterprise teams should pay attention to.
The notice problem
The argument buried in the Starbuck complaint that most enterprise teams are not ready for is the notice argument. Starbuck alleges that Google received actual notice that Bard was generating defamatory content about him and continued to serve those outputs.
In defamation law, actual notice can convert a negligence claim into a more serious actual malice claim. In platform liability law, Section 230 of the Communications Decency Act protects platforms from liability for third-party content, but courts have been skeptical about applying that protection to AI-generated content, where the platform itself is generating the output rather than hosting a user's statement. The Starbuck case adds a layer: if a company receives notice that its AI is generating false and defamatory content about a specific person, and takes no action, the notice itself becomes evidence.
This creates a direct parallel to what enterprise teams face when deploying AI tools. If your AI support chatbot, your AI-powered profile generator, your AI research assistant, or any other tool generates a statement about a named person that is false and harmful, the moment someone reports it to you is the moment your legal clock starts running. What you do in the next 24 to 48 hours matters.
Most enterprise teams do not have a written procedure for this scenario. They have bug report workflows. They have content moderation queues. They do not have a specific escalation path for "our AI generated a false statement about a named individual and that person has now contacted us." That gap is the policy update this case demands.
Three things to update before this becomes your problem
The Starbuck ruling does not change the law overnight. It is one trial court denying a motion to dismiss, not a Supreme Court ruling. But it signals the direction of litigation, and enterprise teams that are ahead of it will be in a better position than teams that treat it as an edge case.
Update 1: Build an AI hallucination complaint procedure. Write down what happens when a named individual contacts you with a claim that your AI generated a false statement about them. Who receives the complaint? Who in legal needs to be notified within how many hours? Who has authority to attempt to suppress or correct the output? What gets documented and where? This procedure should exist before the first complaint arrives, not after.
Update 2: Review your content review policy for AI-generated profiles, bios, and summaries. Any AI tool that generates text describing a specific named person -- a bio generator, a prospect research tool, an automated meeting prep assistant -- is generating content that could be defamatory if it is false. Review whether you have any quality check or review step for outputs about named individuals. If the answer is no, consider what threshold would trigger human review before AI-generated content about a person is displayed to others.
Update 3: Do not assume Section 230 covers your AI outputs. Section 230 protection applies to platforms that host third-party content. Courts have been inconsistent about whether it applies to content an AI generates, as opposed to content a user posts. Some courts have found that AI-generated outputs are more like editorial decisions than hosted user content. Do not design your risk posture around Section 230 coverage that may not exist for AI-generated statements. Ask your legal team specifically about this gap.
Why this case is different from prior AI liability debates
The debates about AI liability so far have mostly focused on products liability (does your AI cause physical harm), copyright (did your AI train on protected content), and consumer protection (does your AI mislead users). The Starbuck case is squarely in defamation territory, which has its own doctrine and its own standards.
Defamation has real teeth. It has a well-developed body of case law. It does not require a regulatory investigation to trigger liability -- a single affected individual can bring a claim. And the damages can be substantial: Starbuck is seeking over $15 million over allegations involving a single chatbot.
The case also involves a relatively common AI failure mode: generating false biographical information about real people. This is not a rare edge case. Anyone who has probed AI chatbots on this topic has encountered it. The question the Starbuck case is now forcing into litigation is whether the companies behind those systems are legally responsible for the harm.
The answer is not yet determined. But the Delaware Superior Court has said the question is worth asking in court.
Related Reading
- FTC AI Enforcement Actions 2026: What the Pattern Means for Small Teams
- Vetting AI Tools: Avoid Fake Apps, Malware, and Data Risk
- Workday AI Lawsuit: HR Screening Checklist 2026
- AI Data Privacy for Small Teams: GDPR and CCPA Guide
- ChatGPT Product Liability Lawsuits JCCP 5431: What Enterprise Teams Need to Know
Sources: Reason / Volokh Conspiracy, July 24, 2026, Bloomberg Law, July 24, 2026, Court opinion PDF via Reason.
