What happened?
In a ruling published on February 11 (case no.: 5 Gs 19/26), the Reutlingen District Court rejected a request by the public prosecutor's office to issue an arrest warrant. The charge was based on a match generated by a facial recognition system operated by the German Federal Criminal Police Office (BKA), which linked a suspect to a man already known to police. The court deemed this algorithmically generated indication insufficient to justify pretrial detention.
The details
The case originated from an incident at a drugstore in October 2025. Store employees observed via video surveillance a person stealing several bottles of perfume. When the suspect entered the store again and was confronted, the situation escalated: while fleeing, he swung an umbrella and struck two employees who tried to restrain him.
Police used the video footage to conduct a facial recognition search through the BKA. The system returned a match: a man already wanted for other offenses. Based on this match and a blanket classification as "previously known to police for relevant offenses," the public prosecutor's office requested an arrest warrant for robbery-like theft (räuberischer Diebstahl).
The court described the software used as "dubious" and criticized the fact that neither its functioning nor its algorithm, the reference data used, nor its error rates were documented in a comprehensible way. The investigators' mere assertion that it was an "improved" piece of software was not sufficient for judicial evaluation of evidence.
The judges also noted deficiencies in investigative rigor: there was no photo lineup in which witnesses could have identified the suspect among several photographs. Objective evidence such as DNA or an analysis of cell tower data was also missing, as was an anthropological expert opinion. The attempt to support the urgent suspicion by referring to the defendant's "prior record" was rejected by the court as an impermissible form of "labeling" that cannot substitute for case-specific facts.
The arrest warrant also failed due to its legal classification: under Section 252 of the German Criminal Code, robbery-like theft requires that the perpetrator be caught "in the act" and use force to retain the stolen goods. Since the person had left the store in the meantime and it was unclear whether he still had the perfume on him when he re-entered, the court did not consider the intent to retain possession sufficiently established.
Context
The BKA operates the official police facial recognition system (GES) for identifying individuals. Since September 2024, an AI-based system has been in use. In 2025, the system recorded around 343,856 search runs — more than double the number from the previous year. These figures illustrate how deeply automated facial recognition has become embedded in investigative practice.
The Reutlingen ruling sets a clear legal boundary for this development: a software match alone, without comprehensible technical documentation regarding the algorithm, reference data, and error rates, is not sufficient to justify as serious an interference with fundamental rights as pretrial detention. For operators of video surveillance systems and security officials in retail, the case shows that camera footage can indeed provide important evidence, but automated analysis is no substitute for traditional investigative work.
Practical tips
- Operators of video surveillance systems in retail should document footage in a way that also supports traditional identification procedures such as photo lineups.
- Security officials should be aware that automated facial recognition alone does not provide a court-proof evidentiary basis — supplementary evidence remains indispensable.
- For criminal defense purposes, the ruling provides a basis for argumentation: when clients are incriminated by algorithms, validation data and quality evidence for the software used can be demanded.
- Companies using or planning to use video surveillance technology with facial recognition functionality should keep an eye on transparency requirements for AI systems, as these are increasingly being scrutinized by courts.
Outlook
The Reutlingen District Court's ruling is likely to have an impact beyond this individual case. Given the sharp increase in the use of facial recognition systems by German authorities, it is expected that other courts will impose similar transparency requirements on AI-supported investigative methods. For investigative authorities, this means that in the future they will need to provide more detailed information on the functioning, error rates, and validation of the software they use if algorithmic matches are to serve as the basis for serious interferences with fundamental rights. Details on possible adjustments to investigative practice or legal regulations are not yet known.
