Baidu in court: AI hallucinations are no grounds for a fine

Imagine spending years building a career in law, only to go online one day and discover that a popular neural network ties you to a series of crimes. That is exactly what happened to Beijing lawyer Huang Guigeng, who decided he would not tolerate the antics of algorithms. He sued Baidu, demanding one million yuan for damage to his professional reputation. The first hearing took place recently in the Haidian District court, and the arguments of Baidu's defense make you wonder what kind of brave new world we live in. The company does not deny that its product produced falsehoods, but says this was not a deliberate act.

Baidu's main argument revolves around the term "hallucinations." The company's lawyers said the phenomenon is an unavoidable stage in the development of any large language model. By their logic, since the technology is still being refined, the developer should not be held responsible for the errors it makes during this "transition period." Baidu insists it had no intent to insult the lawyer, and therefore there is no basis for liability. It is a textbook example of corporations trying to shift responsibility from the actor to the imperfection of the tool, pretending that AI is some elemental force that cannot be fully controlled.

Context matters here. China is currently at the forefront of AI regulation, trying to balance support for its tech giants in the race with the US against tight control over information. If the court sides with Baidu, it will set a dangerous precedent. In effect, any company could put a raw product on the market that generates defamation and cite the technical complexity of the algorithm as its excuse. On the other hand, if Baidu loses, the industry faces a wave of lawsuits. Every wrong word from a chatbot could become grounds for multimillion payouts, which would push developers to "tighten the screws" and make models as cautious as possible and, most likely, less useful.

It is interesting how Baidu tries to separate the technology from responsibility. It claims it committed no direct or indirect infringement of rights, since text generation is an automatic process. But this leads to a logical dead end. If a company profits from the algorithm and promotes it as a reliable source of information, why should it be absolved when the algorithm gets things wrong? In traditional media, an editor cannot say a false article was a "hallucination" of the printing press. But in the world of LLMs the old rules somehow stop working, or at least companies badly want us to believe so.

The Beijing case is only the first swallow in a series of coming legal battles. While OpenAI and Google fend off copyright lawsuits, Baidu faces a more down-to-earth but no less important problem: protecting the honor and dignity of an ordinary person. The outcome will show how "smart" and "irresponsible" artificial intelligence is allowed to be in modern society. If hallucinations are recognized as a machine's legal right to err, we should all brace for a digital reality that is even more unpredictable and toxic.

The key question: Will "hallucination" become a legal term that absolves liability, or will Baidu have to pay for every false token?