The case began with a murder conviction in Hamburg in 2021, secured partly on the basis of an AI-generated facial recognition match that placed the defendant at the scene. When the defence requested the source code and training data behind the matching algorithm, the software vendor refused on commercial confidentiality grounds. The Bundesgerichtshof has now ruled, unanimously, that commercial confidentiality cannot override a defendant's constitutional right to challenge the evidence against them.
The ruling establishes three mandatory requirements for AI-generated forensic evidence in German criminal courts: full disclosure of the algorithm's training data composition and known error rates; an independent technical audit conducted by a court-appointed expert with access to the underlying system; and a written explanation of the specific output in terms accessible to a lay jury. Where any of these cannot be provided, the evidence is inadmissible.
The practical implications are immediate. German prosecutors have identified 847 pending cases in which AI-generated evidence plays a material role, and the Federal Public Prosecution Service has announced a review of all convictions since 2018 in which such evidence was used without the transparency standards the court now requires.
Legal scholars at Humboldt University note that because Germany is the EU's largest jurisdiction, the ruling will effectively set the floor for AI evidence standards across member states ahead of the EU AI Act's criminal justice provisions entering force in 2026.
"An algorithm is not a witness," said presiding judge Dr. Petra Hofmann. "It must be examined like one."