German Court Holds Search Engines Liable for AI-Generated Errors as New AI Law Takes Shape
14.06.2026 - 00:23:49 | boerse-global.de
A landmark ruling from the Landgericht München I at the end of May 2026 has tightened the legal risks for operators of AI-powered search services. The court decided that AI-generated summaries count as the operator's own statements, stripping away the liability privileges that normally protect pure search engines.
The case (file number 26 O 869/26) stemmed from two publishers who sued over false AI-generated summaries that wrongly linked them to fraudulent business practices. The chamber found the operator fully liable for those inaccurate fact claims. While the verdict is not yet final, industry observers note that with billions of daily queries, even a tiny error rate produces millions of mistakes statistically. For platform operators, the cost of getting it wrong just went up dramatically.
New Supervisory Body for AI Takes Shape
Just days before that court decision, the German Bundestag passed the national implementation law for the EU AI Regulation on 11 June 2026. The "KI-Marktüberwachungs-und-Innovationsförderungs-Gesetz" (KI-MIG), backed by the CDU/CSU and SPD coalition, designates the Bundesnetzagentur (Federal Network Agency) as Germany's central AI market watchdog.
The agency will handle citizen complaints, advise businesses, and operate at least one AI regulatory sandbox. To do all this, 43 new positions have been allocated. Annual running costs are estimated at roughly €15.9 million for the federal government and €33.1 million for the states, with a one-time investment of around €4 million. Violations can draw fines of up to €50,000.
Opposition parties and industry associations, including Bitkom, warn the law will create bureaucratic fragmentation because responsibilities vary between federal states. The coalition counters that the legislation provides legal certainty for companies operating in the AI space.
Open-Source Legal AI Reaches Milestone
While lawmakers and regulators move, the developer community has been busy. In mid-June 2026, the GitHub repository "claude-fuer-deutsches-recht" hit version v312.0.0. It now contains over 20,000 specific skills and more than 200 test case files covering areas from labour and corporate law to data protection.
Written in Python and HTML and released under an open-source license, the collection is described by its creators as experimental. They stress it is not a substitute for professional legal advice. Still, the project shows how private AI development is racing ahead even as the state tries to build guardrails.
Privacy Watchdogs Flag Shadow AI and Tax-Data Risks
On 12 June 2026, the state data protection officer for Lower Saxony issued a public warning about employees using private AI accounts for work tasks. This "shadow AI" practice can lead to data protection violations and the leakage of sensitive company information. The recommendation: clear official policies and approved, vetted AI interfaces.
At the federal level, Louisa Specht-Riemenschneider, the Bundesbeauftragte für den Datenschutz, criticised plans to train AI models using real tax data. She called for specific legislation and strict data minimisation to prevent models from memorising personal information. Any such data processing, she argued, must take place exclusively on secure servers operated by the tax administration.
