The German Federal Court of Justice ruled on 13 October 2026 that the legally required formative contribution of the lawyer may be missing when a revision brief is drafted with AI without substantive review. In this case the brief remained admissible, but lawyers must now demonstrably assess the content themselves, check sources and take full responsibility before filing.
What exactly did the German Federal Court of Justice decide about an AI-drafted revision brief?
In decision 3 StR 350/26, the German Federal Court of Justice held that, under §345(2) StPO, a revision brief requires a formative contribution from the defence counsel, who thereby takes responsibility for the entire content. The court said expressly that this contribution may be missing in an individual case when a lawyer adopts an AI-generated set of grounds without substantive review. In this specific case, however, the court did not consider the doubt weighty enough, and the grounds remained valid.
What is important, therefore, is what the BGH did not say: there is no automatic invalidity of every AI-supported court document. The ruling is a warning, not a prohibition.
Why can the lawyer's required contribution legally be missing when AI is used?
The legal basis lies in §345(2) StPO: on behalf of the accused, the revision applications and their grounds may be filed in a document signed by the defence counsel or lawyer, or at the court registry. Against this procedural background, the standard described by the BGH requires the defence counsel to contribute formatively to the grounds and to take responsibility for the entire content. The BGH considers that this formative contribution may be missing in an individual case when a revision brief has been drafted using AI without substantive review by the defence counsel. A tool that mainly produces the text does not carry that responsibility; only the lawyer can.
The open question the BGH leaves is exactly when the contribution is "substantive" and "formative" enough. Our analysis: the ruling makes it all the more important for lawyers to be able to demonstrate that their legal assessment was more than placing a signature under generated text; the ruling does not, however, formulate a general rule that the burden of proof actually shifts.
What does this ruling mean for lawyers, firms and legal decision-makers?
Our analysis: because the BGH ties the lawyer's contribution to the content, a drafting error combined with insufficient human review can grow into an admissibility risk; therefore, in our assessment, firm managers should record that every AI-drafted set of grounds has demonstrably been assessed by a qualified lawyer before filing. Because the ruling does not further define the term "substantive review", in our assessment uncertainty arises that may turn against the lawyer if the file is later contested; therefore, in our assessment, practice group leaders would be wise to formulate an internal standard for what that review entails at a minimum. Because parallel professional rules draw the same line, in our analysis this is not a purely German problem: the CCBE guide on generative AI for lawyers underlines the need to understand and verify AI output, while the SRA warning states that regulated professionals remain responsible and cannot rely on AI output as a defence for errors; therefore, in our assessment, compliance officers should align their policy across multiple jurisdictions, not only with §345 StPO. In our analysis, a demonstrable chain of review from generation to release is the most defensible way of working, because the lawyer can then substantiate how the grounds fit the specific file. The practical consequence, in our assessment, is that an AI court document only holds up when the human involvement is substantively visible and does not exist merely as a signature; anyone who fails to arrange this runs, in our assessment, the risk that an otherwise strong revision founders on a formal defect. This aligns with the broader lesson that human oversight of AI advice must be measurable, and with the question of how law firms deploy AI without steering on outcomes.
Which review steps make an AI-supported court document defensible?
A defensible way of working separates three things that often run together: generating text, editing it and taking legal responsibility. The steps below are our editorial translation of the BGH logic and the supplementary professional guidelines into everyday practice.
- Classify the use. Record whether AI only provided draft text or also formulated legal grounds; that distinction determines how heavy the review must be.
- Check every source. Test each cited ruling, statutory provision and fact against the original source. The practice direction of the Irish Court of Appeal states explicitly that a second AI tool is not verification.
- Assess the fit with the file. Connect the grounds to the specific judgment and remove generic or irrelevant passages.
- Have a qualified lawyer independently edit and release. As an editorial recommendation: have a qualified lawyer independently edit and release the document; the release should be a substantive assessment and not a formality.
- Document the chain of review. Keep a record of who checked and approved what, so that the contribution can later be demonstrated.
- Correct or report errors immediately. If an error is discovered after filing, remedy it as quickly as possible and inform the court where necessary.
Anyone who structurally secures these steps treats AI as a tool and not as a replacement for the legal assessment. This aligns with the broader theme of using AI responsibly in professional legal practice and with the question of how to check AI answers with source verification and an audit trail. The core of the BGH decision is modest but sharp: responsibility remains with the lawyer who formatively drafts, reviews and signs the grounds.
Sources and references
Sources: The article relies on the decision of the German Federal Court of Justice in 3 StR 350/26, §345 StPO, the CCBE guide, the practice direction of the Irish Court of Appeal and the SRA warning.