The Lawsuit AI Invented Cost a Real Company Real Customers

2026-08-25

Last spring, a homeowner in Minnesota picked up the phone to cancel a $150,000 solar installation contract. His reason was specific: he had searched the installer's name, and the AI Overview sitting above Google's results told him the company was "currently facing a lawsuit from the Minnesota Attorney General" over deceptive sales practices.

The company, Wolf River Electric, was not facing that lawsuit. It never had been. The state's attorney general, Keith Ellison, had sued four solar-lending companies the year before over $35 million in concealed fees. Wolf River was not one of them. Google's AI Overview cited four sources for its claim anyway — a Star Tribune article about the real case, an Angie's List listing, a press release from the Attorney General's office, and one more. None of the four said Wolf River had been sued. The system built a false regulatory record out of real, checkable sources and presented it with the confidence of a settled fact.

Wolf River sued Google in March 2025, seeking $110 million to $210 million in damages. The case bounced to federal court and back after Google missed a filing deadline, and as of this year it is still working its way through Minnesota state court. It remains one of the clearest tests yet of a question courts have barely started to answer: what happens when an AI system does not just get an opinion wrong about a company, but fabricates the company's legal status.

Real sources, false synthesis

Most of the AI defamation cases getting attention involve individuals — a chatbot inventing a criminal accusation, a fabricated embezzlement charge, a made-up connection to a riot. Wolf River is a different, and in some ways more dangerous, category of failure. Nobody had to invent a source from nothing. The system took four real, publicly available documents, none of which mentioned the company, and synthesised them into a specific, false, and highly damaging legal claim — then attached citations that made the claim look verified rather than invented.

That distinction matters because it is exactly the failure mode people are least equipped to catch. A wholly fabricated source is at least theoretically checkable: does this article exist? A false synthesis of real sources passes a much lower bar of suspicion. The links resolve. The publications are legitimate. The Attorney General really did sue solar companies that year. Everything checks out except the one sentence that mattered, and by the time a customer is on the phone cancelling a contract, nobody is going back to read the source documents word for word.

The claim outlives the correction

Google has since defended the case partly on the grounds that AI Overviews are experimental and that responsibility for verifying claims sits with the user — an argument that is getting a harder hearing from courts and regulators with each new ruling. But the more immediate problem for any company in Wolf River's position is procedural, not philosophical. By the time the false claim surfaces in a lost contract or a declined partnership, the AI Overview that produced it may already have changed, been corrected quietly, or vanished from that particular search entirely. What is left is a customer's account of what they read, a company's certainty that the claim was false, and very little in between that a court, an insurer, or a journalist can independently check.

That gap is the real story here, more than the size of the damages claim. Wolf River had the advantage of a customer willing to describe exactly what he had seen and when. Most companies encountering a fabricated regulatory claim will not be that lucky — they will hear about it secondhand, weeks later, after the version of the internet that produced it has already moved on. The claim does not need to persist for the damage to be permanent; it only needs to have existed once, unrecorded, at the moment someone was deciding whether to sign a contract.

Regulatory status is a specific, checkable fact — a company either is or is not named in a lawsuit. When an AI system gets that wrong, it is not offering an opinion that reasonable people could dispute. It is stating something false with the appearance of official record. Companies that treat this as a hypothetical are betting that no customer will ever act on what an AI system tells them about their legal standing. Wolf River is the reminder that some of them already have.