Judge Meghan Adams of the Delaware Superior Court on July 24 denied Google's motion to dismiss Robby Starbuck's defamation lawsuit over statements generated by Google's AI systems, sending one of the first US generative-AI defamation cases toward discovery. The opinion, published by the Volokh Conspiracy, recounts Starbuck's allegation that Google AI outputs made 13 categories of false claims about him, including fabricated accusations of sexual assault, criminal convictions, and participation in the January 6 Capitol riot. Google conceded the statements were false but argued they were not actionable.

The court declined Google's lead arguments at this stage. Google contended that AI outputs generated in response to user queries were not "published" to third parties in the defamation sense, and that its experimental-AI disclaimers defeat reliance. Judge Adams found the complaint's allegation that millions of users interacted with the tools enough to plead publication under Delaware's conceivability standard, and declined to weigh disclaimers that sit outside the complaint. The opinion also let Starbuck's actual-malice theory stand for now: his lawyers had sent Google legal notices about earlier false outputs, and the complaint alleges the false statements continued. Bloomberg Law and Law.com's Delaware Business Court Insider covered the ruling.

A denied motion to dismiss decides nothing on the merits. Starbuck, whom both sides agree is a public figure, still has to prove Google acted with knowledge of falsity or reckless disregard for the truth. What changed is that a US court declined to treat AI-generated text as categorically outside defamation law, and pretrial discovery into how the outputs were produced now proceeds.

The takeaway for brands: what AI models assert about a company or a person is starting to carry legal consequences, not just commercial ones. The ruling also gives notice and correction real weight, since the complaint's strongest surviving allegation is that Google was told about the false outputs and they kept coming. Knowing what models say about you, and putting providers on notice when it is false, just became more than reputational hygiene.