Facts here are almost boringly ordinary, which is the point. A cosmetic surgery clinic ran an AI chatbot on its website to schedule appointments and answer routine questions. Asked about the qualifications of the practice's two managing directors, the bot replied that they were "specialists in plastic and aesthetic surgery" and "specialists in aesthetic medicine." In German medical practice those are protected professional titles, earned through a defined training path. The two doctors did not hold them.
Nobody at the clinic wrote that sentence. Nobody approved it. It was generated on the fly, in a chat window, in response to a question a prospective patient asked at some hour when everyone who works there was asleep. The Higher Regional Court of Hamm ordered the clinic to cease and desist anyway.
The Legal Move That Matters
Judges in Hamm did not reach for a novel AI statute, because it did not need one. It deemed the chatbot's incorrect responses to be misleading commercial acts by the company itself, under Section 5(1) and (2)(3) of the German Unfair Competition Act, the UWG. That is the same law that governs a misleading brochure, a misleading price sign or a misleading claim in a television ad.
Strip away the technology and the reasoning is almost mundane: the clinic put a thing on its website that makes commercial statements to customers, that thing made a false statement about the clinic's own credentials, and false statements a business makes to customers about itself are the business's responsibility. Whoever runs a chatbot on their own site carries the risk that it hallucinates.
That quote is two years older than the Hamm ruling and it is the reason the Hamm ruling was predictable. In February 2024 the British Columbia Civil Resolution Tribunal handled a passenger who had been told by Air Canada's chatbot about a bereavement fare policy that did not exist. Air Canada's defense was, in substance, that the bot spoke for itself. The tribunal found that submission remarkable and ordered the airline to pay 812.02 Canadian dollars, which is roughly $570. Small money, large idea. Hamm is the same idea arriving in European unfair-competition law with an injunction attached instead of a refund.
This pattern shows up outside courtrooms too. In April 2025 the AI coding startup Cursor watched its own support bot, named Sam, invent a single-device subscription restriction and tell paying users about it. Co-founder Michael Truell went to Reddit to state plainly: "We have no such policy." A company that builds AI tools for a living could not stop its own AI tool from writing policy for it in public.
Meanwhile, In The Courtroom Itself
The commercial-speech question is only half of the liability story. The other half is what happens when hallucinated text gets filed with a judge, and there the numbers have gone somewhere absurd.
A public database maintained by Damien Charlotin, which catalogues court decisions where AI-hallucinated material showed up in filings, stood at roughly 200 cases in mid-2025. It hit 719 by January 2026, 1,227 by early April, 1,458 on May 22 and 1,598 on June 9, which works out to just under eight new cases a day. As of April 2026, 811 of them were in United States courts, with additional cases documented in the United Kingdom, Singapore and Argentina.
Penalties stopped being symbolic somewhere along the way. The record monetary sanction sits at roughly $109,700 in Couvrette v. Wisnovsky in the District of Oregon in December 2025. Whiting v. City of Athens produced $15,000 per attorney. United States courts imposed more than $145,000 in AI-filing penalties in the first quarter of 2026 alone. Courts have also gone past money into revoked pro hac vice admissions, referrals to bar regulators and personal liability for supervising partners.
One case captures where this is heading better than any tally. Withers v. City of Aberdeen, decided in the Northern District of Mississippi on June 8, 2026. Both sides filed fake citations. Not one careless lawyer against a diligent opponent who caught it, which is the shape of most of these stories. Both of them. The judge canceled the trial outright and suspended the two lead attorneys from practicing in the district for two years.
The Market Already Priced This In
Insurers and regulators moved before most companies did. Lloyd's of London launched hallucination insurance through the startup Armilla in May 2025, which means underwriters have been quoting a price on this exposure for over a year. FINRA flagged hallucinations as a compliance concern for broker-dealers in its 2026 Annual Regulatory Oversight Report. And in June 2026 a company called Scaled Cognition raised $100 million specifically to build hallucination controls. When there is an insurance product, a regulatory flag and a nine-figure funding round all pointing at the same failure mode, the failure mode is not an edge case anymore.
The Honest Complication
Two things should temper how far anyone runs with the Hamm decision. First, it is a German ruling under German unfair-competition law. It binds no court in the United States, and its reasoning is specifically about commercial statements a business makes about itself, which is a narrower category than "anything a chatbot says." A support bot giving a wrong shipping estimate is not automatically the same legal problem as a clinic inventing medical credentials.
Second, the 1,598-case figure comes from a volunteer-maintained public database, not from an official judicial statistic. It counts decisions somebody noticed and logged. The real number is presumably higher, and the growth curve partly reflects better tracking rather than purely worse behavior. Those are honest caveats and they do not change the direction of travel, which is that courts on two continents have now been asked who owns a hallucination and have answered, repeatedly, that it is whoever put the thing in front of the public.
The Verdict
The Higher Regional Court of Hamm ruled on May 12, 2026 that a clinic's website chatbot inventing specialist credentials for its own doctors was misleading commercial conduct by the clinic under the German Unfair Competition Act. Combined with Air Canada's failed 2024 argument that its bot was a separate legal entity, and 1,598 tracked hallucination cases carrying sanctions up to roughly $109,700, the "the AI did it" defense is closing. If you deployed it, you said it.