Why bother searching through link lists in Google when my favorite AI can give me a plausible-sounding answer in just a few seconds? Maybe because AI answers are often error-prone. But who is actually to blame if my health is worse because I listened to an AI – A commentary analysis.
AI liability: OLG Hamm makes groundbreaking ruling
- The Hamm Higher Regional Court made a groundbreaking ruling in May 2026. Accordingly, companies that integrate AI services in the form of chatbots on their websites are liable for the statements made – even if they are fictitious. So anyone who relies on AI in customer support will take care of it full risk.
- In this specific case, Aesthetify GmbH, which offers beauty services, had integrated a chatbot for customers. He was used to book appointments and was able to answer questions. The problem: In the underlying chat, the bot claimed that the two managing directors have specialist titles that, firstly, they do not have and, secondly, some of which do not even exist in Germany. Blunt lies so.
- At the same time, a study by the Baden-Württemberg State Media Center found that around half of the young people surveyed between the ages of 12 and 19 Information from AI chatbots is fundamentally familiar. As part of the “News Integrity in AI Assistants” study, the European Broadcasting Union came to the conclusion that 45 percent of all AI answers have at least one serious weakness – i.e. incorrect or outdated information. Trust in things and people who don’t tell the truth is increasing – US President Donald Trump also shows this.
Why AI fails twice – and no one intervenes
The case before the Hamm Higher Regional Court is no longer an isolated case. There are an increasing number of cases around the globe – primarily civil law – concerning who is liable for the statements made by AI chatbots. The background is that it more and more crimes There are cases in which AI was demonstrably involved.
AI acts doubly incorrectly. On the one hand, programs like ChatGPT can currently be used too easily persuadedto reveal dangerous information. Often all you have to do is pretend to be researching a book or film and all security precautions are dropped. The result: More and more (suicides) murders have been and are being guided by artificial intelligence.
On the other hand, ChatGPT, Grok and the like don’t trade enough. At the latest when the first question comes about life-threatening drug use, building a weapon or tying a rope, not only the chatbot would have to end the conversation immediately. It would also have to internal alarm goes off. The system would have to alert human examiners to people who are potentially at risk and – if possible – to local authorities about people’s acute risk.
Building such a safety net to protect your own users should be one of the minimum requirements for AI systems. Instead, the energy has so far migrated in unbelievable excuseswhen it is already too late and someone has died.
Voices
- Marius Pflaum from the European law firm network DIRO believes that the Hamm Higher Regional Court got the ball rolling with its ruling. He comments: “The decision has considerable implications beyond the medical sector. Companies that use generative AI systems in their external presence – for example for advice, product information, contract information or support – must ensure that the content generated is continuously monitored and technically limited. Companies cannot hide behind the supposed autonomy of generative systems. Anyone who uses AI as the mouthpiece of their company must accept their statements as their own explanations. This applies even if the AI model has been properly trained.”
- Christian Strohmayr, AI and content expert and managing director of BEWEGT.content GmbHbelieves that a tension is emerging between technology and law. German hostility to technology could become a disadvantage for the location. He writes on LinkedIn: “An AI system is legally part of the company organization. Companies are therefore fully liable for statements generated by their AI. This may be legally understandable. However, it could be problematic economically. Germany is developing from a location for innovation to a location for regulation. If companies have to be afraid of using AI at all, international competitiveness will suffer massively.”
- Fabian Peters, editor-in-chief of BASIC thinkingcommented on the responsibility of AI operators in the “Break The Week” podcast. Specifically, it’s about a murder in Florida, in which ChatGPT is said to have given detailed instructions on how to build weapons: “When I ask a search engine like Google, it doesn’t provide as detailed answers as ChatGPT. If you ask ChatGPT about weapons or attack tactics and the thing spits it out, I see OpenAI as having a certain responsibility because such problems can be solved simply through programming. In other words, there are simply no answers for certain keywords.” He adds: “The jurisprudence sometimes cannot keep up because technological progress is too fast for the judiciary.”
How AI liability could change in Europe
Theoretically, there are definitely starting points for making AI responsible for its statements. One starting point is, among other things, European product liability law. Accordingly, companies are liable for their products, just as we know from the playground or the zoo: Parents are responsible for their children.
If a customer is harmed by a product – and this raises the question of whether an AI chatbot is a product – companies are (possibly) liable. When machines have design defects and as a result Human lives are at riskthe machine manufacturer is responsible for this. So when ChatGPT develops murder instructions or xAI creates naked pictures of children and this causes physical or psychological harm, OpenAI or Elon Musk have to hold their heads.
In order to force AI companies to take on more responsibility, it is important to tighten the applicable rules. So far it will Personal responsibility is largely passed on to the users. Anyone who accepts health tips from an AI has only themselves to blame. After all, there is evidence everywhere that AI can make mistakes. But that’s not enough.
One solution could be to not look at the general purpose in case law – for example, the search for information. Instead, AI systems should Worst case scenarios be assessed. Does an AI potentially give out life-threatening information? Then the liability risk must also increase accordingly. Anyone who drives a Ferrari also pays higher insurance amounts than Polo drivers.
Also interesting: