Report from The Decoder
In Brief – The German Regional Court of Munich has ruled that Google is directly liable for false statements produced by its AI Overviews feature. The case involves two plaintiffs who alleged that Google’s AI summaries wrongly linked their Munich-based publishing companies to scams. The court determined that the AI-generated overviews constituted Google’s own content and rejected the liability protection given to search engines that produce links to third-party content. The judges rejected Google’s argument that users are informed that the AI answers are not always accurate and that they should verify information on their own, and they noted that the language of Google’s AI summaries are sometimes not backed up by the sources that Google links in their AI-generated response. The court also found that AI-generated responses deserve less free speech protection because they reflect algorithmic outputs rather than human convictions.
Context – AI chatbots and search are converging. And Generative AI chatbots “hallucinate”, meaning they sometimes formulate responses that are completely incorrect. AI designers don’t fully know how their systems work and so can’t truly stop hallucinations. This has bedeviled European regulators. When Italy’s data protection authority banned OpenAI for a few months in 2023 they raised the concern that hallucinations violated the GDPR’s principle of data accuracy. The European Data Protection Board Taskforce on Chat-GPT’s 2024 report also detailed the problem and suggested OpenAI provide users with an “explicit reference” that generated text “may be biased or made up” and that the “probabilistic output” of the chatbot has a “limited level of reliability”. Giving clear warning that AI responses might be in error is at the heart of the key hallucination liability case in the US, a defamation lawsuit in Georgia. The judge dismissed the lawsuit against OpenAI and cited the company’s “extensive warnings to users that errors of this kind could occur.” If German law goes the opposite direction, as the Munich court did in rejecting Google’s defense, GAI won’t be tenable in the country.