Google's AI invented a sponsorship that doesn't exist — and reran the lie 2 times out of 25

A confident, repeatable, randomized falsehood

The original complaint is small and exact: Google's AI stated that repair-rights advocate Louis Rossmann is sponsored by Ground News. He isn't. What turned it into a discussion was the statistics of the error. One reader ran the exact query 25 times: “it generated a different response each time, and 2/25 times it said he was sponsored. So … Google doesn't have any kind of 'lock in' for facts.” That non-determinism is the whole problem — another reader reported googling his own company and getting “at least one factual error, and … the errors were different for each run.” You can't correct a mistake that only shows up a twelfth of the time.

The question that turns annoyance into law

The discussion's center of gravity was a pointer from Simon Willison, creator of Datasette, to a Bruce Schneier piece on AI and liability:

“To allow businesses to hide behind the excuse of faulty AI … would be a massive handout to companies … Why hire human writers, lawyers or doctors when AIs are not only cheaper, but also absolve employers whenever they make a mistake?”

One reader put the legal question plainly: “Could LLMs libel an individual? … Who would that person sue?” With a randomized falsehood and no human in the loop, the chain of responsibility genuinely blurs.

The defense — and why it doesn't hold

The shrug position got stated cleanly: “LLMs hallucinate. It's an unfixable problem. We're going to have to get used to it.” But that's precisely the incentive Schneier warns about. If “unfixable” becomes “not our fault,” a company gets to publish randomized claims about real people at scale and pay nothing when one defames. The fix the thread gestured toward isn't better models — it's refusing to let “the AI said it” be a sentence that ends a lawsuit.