German AI Legal Brief Liability
5 steps1 pieces4 Sources
Germany's Federal Court of Justice (BGH) has established that lawyers must substantively review and take responsibility for AI-generated legal filings, or risk having appeals deemed inadmissible. In a decision dated October 13, 2026 (Case No. 3 StR 350/26), the BGH's Third Criminal Senate ruled that while the specific appeal in question was reviewed on its merits and rejected as unfounded, the use of AI without "substantive control" by the defense attorney can violate Section 345(2) of the Code of Criminal Procedure (StPO). This provision requires active involvement and full assumption of responsibility by the lawyer to protect defendants' interests and prevent court overload.
The case originated from an appeal against a Kaiserslautern Regional Court judgment. The BGH noted that the brief contained standardized passages with little connection to the specific facts, such as abstract discussions of withdrawal possibilities in attempted crimes, ignoring the lower court's finding that police had already been alerted. Although these flaws raised doubts about the lawyer's actual involvement, they were not sufficient to declare the filing formally inadmissible in this instance. However, the ruling serves as a strict warning: uncritical reliance on AI generators creates a significant risk of procedural rejection.
This decision aligns with broader regulatory pressures in Germany and the EU. The EU AI Act, which entered into force in August 2024, classifies AI systems used by judicial authorities as high-risk, requiring human oversight. Additionally, the Berlin Court of Appeal previously reprimanded lawyers for AI hallucinations, including invented case numbers. Lawyers are now under increased pressure to implement rigorous verification protocols for any AI-assisted drafting, ensuring that machine-generated text is critically evaluated and personally adopted before submission.
How it got here
August 1, 2024: The EU Artificial Intelligence Act enters into force, establishing a framework for high-risk AI systems, including those used in judicial contexts.
ibanet.org ↗