Germany de.indymedia
FAQ: What you should know about the behavioral scanner in Berlin
Software evaluates how the people being filmed behave: This form of video surveillance is currently being introduced in Berlin. We clarify the most important questions.
Deutsch (original) ·Auto-translated to English
Software evaluates how the people being filmed behave: This form of video surveillance is currently being introduced in Berlin. We clarify the most important questions.
What is the status of automated behavior analysis in Berlin?
The technology will initially be used with around 30 cameras at Kottbusser Tor. The calibration phase, where camera angles and coverage areas were set up, is complete. The cameras and software have been running in test mode off and on since October 6th. A red light on the camera beams indicates when recording is taking place.
What's next?
After the test, the “real trial operation” follows, in which the system records and analyzes 24 hours a day. The police did not respond to the question of when this would begin. It will probably end in 2028. By then, Berlin's potential coalition partners (Left, Greens, SPD) have agreed to an evaluation of video surveillance, "with the involvement of the police."https://netzpolitik.org/2026/linke-gruene-und-spd-was-ausser-vergesellschaftsc...the exploratory paper.
Accordingly, they apparently expect that AI-supported video surveillance of public spaces will still take place in Berlin in 2028. According to negotiators, the future of AI cameras has not yet been specifically discussed. These are a topic for the coalition negotiations, which are scheduled to begin at the beginning of November. The SPD, which helped introduce it, will defend it, the left wants to have it dismantled, and the Greens are still keeping their position covered.
How does automated behavior analysis work?
Shehttps://www.berlin.de/polizei/polizeimeldeen/fakten-hintergruende/arti...permanentwhat people under video surveillance are currently doing. To do this, software places a kind of stick figure over the people depicted - it describes their poses and movements with vectors in a mathematical space. She then compares the result with her training data. She also looks at the entire video image and tries to assess the context of the movement. If it determines that something looks sufficiently likely to be relevant to the police, it issues a report. A monitor then comes on in the video control center showing the situation. And the police officer on duty decides what happens next.
What options does the police officer have in the video control center?
She can classify the situation as harmless and ignore it - or, for example, send out a patrol to check IDs or even search people. Probably more the latter: a well-studied psychological effect calledhttps://www.bpb.de/lernen/digitale-bildung/werkstatt/569392/automation-b...biasensures that people tend to trust the results of software. At Kottbusser Tor – a designated knife and weapon ban zone – the police are also allowed to carry out checks regardless of the reason. It is not necessary for the person on the screen to recognize a crime; they can also order the check as a prophylactic measure.
What happens in the event of an inspection?
Being stopped by the police, possibly without any reason, is potentially humiliating and can even be traumatizing. Such checks escalate again and again, sometimes with the police using massive amounts of force. In this case, it is also possible that she will report the other person for resistance, which could result in a conviction.
What movements are you specifically looking for?
In Mannheim, where such technologyhttps://netzpolitik.org/2025/behavioursscanner-im-mannheim-hier-wird-die...used since 2018There is a list of 15 behavioral categories that the AI is supposed to distinguish between. These include, for example: hitting, kicking, pushing, choking - the detection of violence is the legal mandate of the software. In addition, it also identifies activities such as dancing, hugging and cycling. This is necessary so that she can differentiate between harmless situations and the ones she is looking for. In Berlin, however, there is no such list. Which movements are being searched for there is a black box.
–––
Against ID checks online.
We are fighting for an open internet.https://netzpolitik.org/spenden/?via=rss">Thisonly possible with your support.
–––
Because the software works with “stick figures,” it is considered privacy-preserving. What's that about it?
Due to the abstraction, the software actually ignores identifiers such as face, skin color or tattoos. However, the movement pattern of a human beinghttps://netzpolitik.org/2025/biometric-gangcognition-show-mir-how-du...unique, People can be identified through this. In addition, the data protection claim is misleading because the full-resolution video image is stored for at least 28 days. So when the software marks a situation and the police want to know who was there as the perpetrator or witness, they take a screenshot and throw it into thehttps://netzpolitik.org/2026/inpol-file-polizeibehoerden-nutzen-immer-o...of the BKA.
If the person concerned has already been processed for identification purposes or has applied for asylum in Germany, their name, date of birth and address are omitted. And with the behavioral analysis, Schwarz-Rot has also given the Berlin police a face search engine. This would allow her to find out where the person can be found on the internet - as soon as the federal and state governments have found a workaround to comply with EU legal requirements when using this technology. Ultimately, the intrusion is no smaller than with conventional video surveillance, but in case of doubt there is a second, equally massive invasion of privacy in addition to classic video surveillance.
Does behavior analysis actually help?
Nobody knows that. An independent evaluation has not been carried out in any of the German cities where the system is currently running.https://marx.wtf/2024/12/24/der-hansaplatz-experiment-war-no-success/">InHamburg revealedA count published by the police shows that around 99 percent of the situations that the AI flags turn out to be false alarms. In Mannheim, even after eight years of using AI, the police cannot name a single situation in which behavior analysis helped prevent or at least solve a crime. The hoped-for deterrent effect tends not to work for emotional acts of violence and potential terrorists; for rational crimes such as drug trafficking or pickpocketing, it can be assumed that they will be pushed into side streets, for example.
Where does the training data come from?
From the perspective of the AI trainer, there is far too little crime in the wild. In Mannheim, AI training is therefore manual work. Police officers have had to go there so farhttps://netzpolitik.org/2026/behaviourscognition-in-mannheim-10-000-film...Show fightscomplete under the cameras. The software is still not in practical use. The Mannheim police assume that 10,000 show fights will be necessary before it runs reliably. Berlin saves itself this effort. The software used here was trained with AI-generated scenes. Software that estimates what crimes might look like taught another software what a crime supposedly looks like. Legally speaking, Berlin is allowedhttps://gesetze.berlin.de/bsbe/document/jlr-NNLBE00004772NN00000000249">AIwith personal datatrain by citizens, even without their consent. The police are even allowed to pass on the data to the AI manufacturing companies for training.
Does this technology discriminate against people?
For example, if a person moves unusually due to physical limitations, the software has a harder time assessing what that person is doing. And since the software tends to produce too many false alarms because otherwise the police run the risk of missing a crime, if they cannot assess a situation, they will probably issue an alarm. This means that people with unusual movement patterns are discriminated against because they are subject to increased surveillance and possibly control pressure. And those who are discriminated against even more directly are homeless people. Because the Berlin software also sounds the alarm when people are lying down - officially to detect “accidents”. In fact, homeless people are also exposed to increased surveillance and control pressure. In addition, this form of surveillance primarily affects marginalized people. It is used where scenes of drinkers or drug users meet. People with money can avoid surveillance. They drink their beer in pixelated restaurants instead of being filmed on the park bench, or they drive cars instead of having to subject themselves to automated surveillance at central transfer stations.
Who developed the behavior analysis software?
Staige GmbH. She is a daughter ofhttps://netzpolitik.org/2026/videoueberwachung-in-berlin-adesso-se-baut-...SE, an international IT holding company based in Dortmund. It is something like the general service provider for the project and subcontracts to Staige for the software and to Axis, a Swedish company, for the cameras. Adesso is known for having relatively extensive access to the company's IT system in 2022/23. It is unusual for a private company to come into play. The behavior analysis in Mannheim and Hamburg was carried out with a Fraunhofer Institute, a non-profit organization.
–––
The work of netzpolitik.org is financed by donations from our readers.
You too can support our public interest-oriented, advertising- and tracking-free journalism.https://netzpolitik.org/spenden/?via=rss">Donatenow..
–––
Where is this already happening?
In Mannheim since 2018 on the station forecourt and on the central shopping mile. Hamburg was added in 2023, where she now works in two places near the train station. Last year Bremenhttps://netzpolitik.org/2026/ueberwachung-im-oepnv-mit-dieser-realsatire...Behavior analysis in tramsinstalled, and since this year they have also been available on public transport in Dortmund and Düsseldorf.
Where else does that come from?
Automated monitoring will be the new normal. The Bundestag has this at the end of Septemberhttps://netzpolitik.org/2026/neues-bundespolizeigesetz-der-bundestag-hat...adopted. This means that the Federal Police can use behavioral scanners at all German train stations. Numerous federal states have already permitted their use in public places or are working on it. In Berlin, at least that was the Black-Red's plan, the Kottbusser Tor would be followed by the Warsaw Bridge, this year also by the Red Town Hall, the Interior Administration, the Jewish Museum, and next year by Alexanderplatz and the Görlitzer Park and the neighborhood around it. It is permitted to be used on all presumed “https://netzpolitik.org/2026/kritikitaetstreib-in-berlin-problemat...polluted places“ and “facilities at risk”. The coalition negotiations from the end of October will show what will actually be achieved.
What is the biggest risk of this form of surveillance?
The pressure to conform. Especially if you don't know exactly what the AI is looking for, people will behave more adaptably. Yet diversity is also the basis of expression of a free and open society - and ultimately of democracy. Unusual behavior expands the scope of thought and possibilities, but with this technology it becomes a cause of suspicion. Automated surveillance means a dehumanization of control.
In what context is this technology being introduced?
A surveillance AI tool rarely comes alone: Together with the behavioral scanner, Schwarz-Rot also approved a face search engine in December 2025, as well as mega databases with automated analysis à la Palantir and video drones, which are also expected to monitor Berlin automatically.https://netzpolitik.org/tag/polizeigesetze/">Similar aspirationsexists in the federal government and in many states. At the same timehttps://netzpolitik.org/2026/geheimdienstreform-zeitenwende-fuer-spione/...the secret servicesGain access to video surveillance systems. And in Baden-Württemberg they are already considering combining behavior analysis with real-time facial recognition.
What can you do against behavioral scanners?
In addition to forms of protest such as demos, information stands or performative theater in the monitored area, there is also the option of camerashttps://netzpolitik.org/2026/mit-kaugummis-laserpointern-und-brecheisen/...to cover up, as long as you don't restrict their function. Anyone who moves within the surveillance area is generally allowed to do sohttps://netzpolitik.org/2024/automatierte-facescognition-wie-das-ve...mask up.
–––
Your donation for digital freedom rights
We report on current internet policy developments, uncover scandals and initiate debates. We are completely independent. Because our fight for digital freedom rights is financed by donations from our readers.https://netzpolitik.org/spenden/?via=rss">Nowdonate.