Germany · de.indymedia · · 1h
Behavioral scanner pilot project in Berlin: “The possibility of anonymous use of public space is disappearing”
Deutsch (original) · Auto-translated to English
We are talking to Berlin's data protection officer Meike Kamp about the automated behavioral scanners that will soon be filming people continuously for no reason at several locations in Berlin. She criticizes the project, but is skeptical about mandatory controls of the new technology by her authority.
An initial four-week pilot project for police video surveillance with automated behavioral scanners begins at Kottbusser Tor in Berlin. Software will then constantly evaluate recordings from video cameras to determine whether predetermined behavior or movement patterns can be found in them. The test will be expanded next year: Alexanderplatz and the Warschauer Brücke will then also be monitored with behavioral detection, because these places are also designated as “crime-prone places”.
Thehttps://pardok.parlament-berlin.de/starweb/adis/citat/VT/19/SchrAnfr/S19...should “probably in the 4th quarter of 2026”start, as can be seen from a response from the Berlin Senate Department for the Interior to a request from the Greens. Things are also set to start soon on the Warsaw Bridge: it is planned to be commissioned “by the end of 2026”. At Alexanderplatz and Görlitzer Park, however, implementation is planned for 2027.
Automated behavioral monitoring will also be available for police property protection around Berlin's Red Town Hall. Tests there and also at the headquarters of the Senate Department of the Interior and at the Jewish Museum are scheduled to start “probably in the second quarter of 2027”.
It is not yet clear whether the projects that are intended to automatically identify dangerous situations will continue unchanged after the election on Sunday in Berlin. It is not yet clear which government constellation will come into being. The party with the most votes in Berlin, the Left, had announced that it wanted to cut back police powers for video surveillance and random checks. Together with the Greens, they took action against the police lawhttps://netzpolitik.org/2026/berliner-polizeigesetz-linke-und-gruene-zie...of the state of Berlin. The Left also promised to abolish the requirement for video surveillance, namely the designation of special crime-prone areas.
https://netzpolitik.org/2026/berliner-polizeigesetz-linke-und-gruene-zie...and the Greens are going to the Constitutional Court
As a service provider for long-term observationhttps://netzpolitik.org/2026/videoueberwachung-in-berlin-adesso-se-baut-...Adesso SE Groupreceived the contract after the tender, the partner is Staige GmbH, in which Adesso is involved and from which the software is supplied. It is clear which unusual behavior should lead to an alerthttps://netzpolitik.org/2026/kameras-und-dronen-in-berlin-was- Behavior...out.
The behavioral scanner cameras have already been installed at Kottbusser Tor. The project will not be cheap for Berlin taxpayers: the Senate Department of the Interior puts the “total investment costs” for this area alone at 1,888,000 euros. On top of that, there should be annual operating costs of 349,000 euros.
We're talking to herhttps://www.datenschutz-berlin.de/">BerlinCommissioner for data protection and freedom of information https://www.datenschutz-berlin.de/ueber-uns/die-rechner/">MeikeCampabout the legal side of automated behavior monitoring and the General Security and Order Act (https://www.gesetze.berlin.de/bsbe/document/jlr-NNLBE00004772">ASOG) – this is the name of the Berlin police law, whose amendment in 2026 made the behavioral scanners possible.
What is a “crime-prone place”?
netzpolitik.org: The pilot project for behavior recognition is intended to permanently film people and analyze their behavior. What data protection issues arise from this? What is the content of the current audit by your authority?
Meike Kamp: I cannot comment on the ongoing process, but regardless of the specific project, from a data protection perspective, the training and data types, the effectiveness of anonymization or pseudonymization procedures, the technical implementation of the training environment and the infrastructure on which an AI system runs play a role when using such AI systems. It also deals with the procedures and classification of detection cases and the legal basis for such cases under police law, detection rates and identification targets, measures for deletion or implementation of further rights of those affected and transparency measures. In addition, the discrimination risks that can arise from the use of an AI system are also relevant.
netzpolitik.org: The project starts in the Kottbusser Tor area, other so-called “crime-ridden places” are to follow. How is a “crime-prone place” defined in Berlin and what restrictions on fundamental rights apply there?
Meike Kamp: What a crime-prone place is and how it is classified is regulated by Section 17a ASOG. Accordingly, this is usually determined by the Senate Department of the Interiorhttps://gesetze.berlin.de/bsbe/document/jlr-NNLBE00004779">Legal Ordinance...and informs the House of Representatives about it. The requirements for classification as a crime-prone location remain low. All that is required is that facts justify the assumption that people there are planning, preparing or committing crimes of significant importance. The Berlin police have a list of the locations on theirshttps://www.berlin.de/polizei/polizeimeldeen/fakten-hintergruende/arti...published.
During the ASOG amendment, I argued that Parliament should decide on this essential issue. Ultimately, the classification of a place as crime-prone forms a basis for the use of video surveillance and other police powers. It is therefore a very important element of the offense that gives rise to the possibility of encroaching on fundamental rights through police measures. This also makes the use of automated behavioral pattern recognition possible.
The crime-prone places are public spaces in the city center that are used by tens of thousands of people every day and sometimes merge into one another. Video surveillance of these places results in a large number of people being recorded who are there without them having given any reason for the surveillance measure. Görlitzer Park, for example, is one of the few green areas in Kreuzberg that is classified as a crime-prone place. People who want to avoid video surveillance would have to forego recreational use in the park.
netzpolitik.org: What would be considered a “success” of software-based video surveillance evaluation in these “crime-prone places”?
Meike Kamp: I don't think in categories like "success", but rather measure police measures based on whether the defined goals and purposes are legitimate, whether the measure is suitable and necessary to achieve these goals, and whether a measure can actually be designed in such a way that it does not disproportionately interfere with fundamental rights overall.
“No further information” about the preliminary test
netzpolitik.org: What data protection regulations are the police bound to in this pilot project in public spaces?
Meike Kamp: In addition to the powers in the ASOG, Section 24e ASOG, in particular the so-calledJI policy. This European directive regulates, among other things, the protection of personal data by security and law enforcement authorities - it is essentially the GDPR for the police.
netzpolitik.org: What data was used to train the software used for behavioral recognition? Were you a data protection officer?https://netzpolitik.org/2026/ki-training-berliner- Behaviorscanner-soll...traininginvolved in an advisory capacity?
Meike Kamp: I can't say anything about that at this point. My authority was not involved in the selection of the software or the training. We will investigate these questions as part of our audit.
netzpolitik.org: The police havehttps://www.rbb24.de/politik/teil/2026/08/videokameras-ki-kottbusser-...above the rbbstated that they had already “successfully” tested the software used on their own property in May. Have the results of this test been made known to you and made available to you? Do you know what type of testing took place and in what test environment the software was tested?
Meike Kamp: I don't have any further information about this yet.
False positive alarms cannot be completely ruled out
netzpolitik.org: What error rates (false alarms) from such tests do you consider to be acceptable if such software is actually used in the pilot project?
Meike Kamp: False positive alarms cannot be completely ruled out. It is important that we always work to reduce the error rate and ensure that - as provided for in the legal basis - an operation is only triggered after a human decision.
The question of the error rate is difficult to answer: The fact that crimes are significantly rarer than comparable circumstances places high demands on the performance of the system used. Error rates that are higher than the probability of crime can lead to a predominance of false alarms.
netzpolitik.org: What data could be used to combine video surveillance with behavioral scanners according to the new regulations in the ASOG? Where do you see particularly problematic aspects?
Meike Kamp: The many references in the legal text make it much more difficult to take an overall view of all powers and their effects in practice. I'm not sure whether even the authors are clear about what data can actually end up in the automated analysis platform or in AI training. Or which powers could be used in one place at the same time - this is of great importance in the area of police powers.
In any case, the new regulations enable the retrograde evaluation of stored image recordings over a period of up to one month, and longer in individual cases. However, the law clearly excludes the use of image recordings for biometric remote identification such as facial or gait recognition.
This, together with the ban on the automated triggering of official measures, is an important limit, but does not prevent event-related identification of individual people from the stored video material. Manual profiling of movements or social contacts in the monitored area, for example, remains technically possible at a later date.
A link with other data via the automated analysis platform may be possible on a case-by-case basis, for example if video recordings are already considered as evidence, but the video recordings and their automated evaluation are not among the types of data that the police may enter there en bloc.
All in all, the police law as a whole lacks an overall, cross-authority perspective that does not just look at individual measures and checks them for constitutional conformity. We would also have to look at how the situation works in reality in a big city like Berlin with tight spatial conditions. I see the danger that large parts of Berlin's city center can hardly be passed through unmonitored and the possibility of anonymous use of public space, which should actually be a given, disappears.
How to achieve more data-saving processes
netzpolitik.org: The legal intervention threshold for automated behavioral monitoring does not differ from that of video surveillance without software evaluation. Do you think this is appropriate and in accordance with the case law of the Federal Constitutional Court, also because an opaque “AI” is trained and allowed to be used?
Meike Kamp: Our test procedure has not yet been completed, so we cannot currently carry out a final assessment of the technology used based on the case law of the Federal Constitutional Court on the use of complex automated evaluation and analysis.
Transparency is an important factor. However, if training is carried out with the help of effective anonymization techniques or, for example, with synthetic data and AI-supported processes would lead to a better distinction between more relevant and non-relevant video material and to display and store it more selectively, this could also lead to more data-efficient processes in the long term. To achieve this, more data-saving investigation methods would have to be the aim of the operation in addition to the police purposes.
netzpolitik.org: Should you as data protection officer require mandatory controls of this technology?
Meike Kamp: The ASOG already provides for mandatory random checks by my authority for certain surveillance powers. I am skeptical about mandatory controls because they can also affect my independence.
What if I am not adequately resourced and I am forced by a legal obligation to have controls that I would have to prioritize over other more intrusive processes for which I then have no resources. In addition, what use are mandatory controls if?https://netzpolitik.org/2026/berlin-gegen-aushoehlung-der-inf...Authorities are not sufficientto prohibit further processing or to impose a fine. Mandatory controls are then written into the law as “compensatory” measures for the encroachments on fundamental rights, although ultimately they are not.
netzpolitik.org: Thank you for the interview!
–––
The work of netzpolitik.org is financed by donations from our readers.
Become part of this unique community and support our public interest-oriented, advertising- and tracking-free journalismhttps://netzpolitik.org/spenden/?via=rss">nowwith a donation.
Read the full story at the source
Source: de.indymedia