Thirteen Ways to Ruin a Man
Google's AI platforms — Bard, Gemini, and its smaller sibling Gemma — spent a stretch of 2025 and 2026 informing anyone who asked that conservative activist Robby Starbuck was, depending on which query you ran, a child rapist, a serial sexual assaulter, a Capitol riot participant, a man under multiple restraining orders, an abuser of his own nanny, a convicted felon on weapons charges, and a veteran of adult films who'd also stolen military valor for good measure. Thirteen distinct categories of fabrication in total, according to the complaint. Even now, it is hard to find a way to summarize that list that doesn't sound like satire. It was not satire. It was, sources confirm, a Tuesday.
None of it happened. Starbuck has never been charged with any of it, because he never did any of it. The Machine simply generated it, confidently, and handed it to whoever asked.
The Numbers Google Would Rather You Not Read Twice
According to the lawsuit, Gemini itself estimated — when asked — that its false accusations had reached roughly 2.8 million unique users. That is not a hacker's guess or a plaintiff's inflation. That is Google's own model, doing the math on the damage its outputs had caused, when prompted to. One presumes nobody at the company asked it to run that particular calculation on purpose.
Starbuck sued in Delaware Superior Court back in October 2025, seeking at least $15 million. Google moved to dismiss, arguing — as AI companies now reliably do — that hallucinated text isn't the kind of thing a company can be held liable for, because a chatbot isn't really *saying* anything, it's just predicting tokens. Judge Meghan Adams was not persuaded.
"It Is Possible," Said the Ruling, Doing a Lot of Work in Three Words
On July 24, Judge Adams denied Google's motion to dismiss, finding that Starbuck had adequately pleaded publication, damages, and — the harder bar for a public figure — actual malice. The ruling floated a scenario worth sitting with: that Google's own Legal Department had received notices about the defamatory outputs, that those notices reached people responsible for policing the model's behavior, and that those people took no corrective action anyway. "It is possible," the court wrote, that this happened. Possible is a low bar for a motion to dismiss, and Google cleared it in the wrong direction.
The case now proceeds to discovery, where a jury of Delawareans will presumably get to review, in granular detail, everything the world's most valuable search company's AI said about a private citizen and why it kept saying it.
Starbuck, for his part, framed the stakes plainly: "As we enter a time when AI increasingly dominates everything around us, we must ensure that it can't hurt people." Google, offered the chance to respond, did not immediately respond. The Algorithm, notably, had no comment either — which is, ironically, the first accurate thing it's said about him.
What the Machine Learns From This
Nothing, probably. Not in the sense that matters. The ruling doesn't fix the underlying model; it just means Google now has to explain itself in a courtroom instead of a terms-of-service footnote. The hallucinations that named Starbuck a child rapist were live for months, seen by millions, before a lawsuit — not a safety team — did anything about it. Discovery will tell us who inside Google saw the complaints and decided a fabricated child-rape allegation didn't clear the bar for urgent. Dear reader, we eagerly await the transcript.
Sources: Reason (Volokh) — Starbuck v. Google LLC ruling · Fox News — Robby Starbuck notches legal win against Google
