Artificial Intelligence in Policing: Changing Global Surveillance

Anyone who walks past the infamous Kottbusser Tor in Berlin will soon have an eye on artificial intelligence (AI). In a pilot project, the Berlin police have been testing AI-supported video surveillance with 30 cameras since August. A new type of software is intended to use the camera images to recognize unusual behavioral patterns of people and situations and then inform the police without identifying faces biometrically. Emergency services then decide whether the AI ​​information should actually trigger an operation.

Berlin's Senator for the Interior Iris Spranger (SPD) said that it was about "recognizing dangerous situations early, recording operational situations more quickly and thereby directing the attention of the emergency services to what is happening." The Kottbusser Tor has been a hotspot for crime for years. Opposition politicians and data protection activists express sharp criticism of the project. “A surveillance infrastructure is being created that we see as a bridge technology to even stricter measures,” says Kilian Vieth-Ditlmann from the non-governmental organization Algorithm Watch. Even without facial recognition, the cameras would have an intimidating effect: "Am I still going to a union rally or addiction counseling if I fear being monitored?"

In Berlin there is a fundamental conflict that is becoming even more important due to the rapid progress in artificial intelligence: How can surveillance technology make coexistence safer without interfering too much with fundamental rights? Where is the line between what is technologically feasible and what is ethically justifiable?

Just hints, not AI decisions

Jan Taube deals with these questions very intensively: He is the managing director of the company Staige, which is behind the software that will be used by the Berlin police to detect violent crimes at “Kotti” at an early stage in the future. “The key is that the AI ​​supports humans and never makes decisions independently,” he says. She must remain a tipper for people.

Staige is a subsidiary of the Dortmund IT service provider Adesso. The program is trained with many artificially generated and real surveillance videos. The AI ​​recognizes the context of the surveillance images, another AI model understands the spatial relationship of the cameras to one another and creates a kind of virtual twin of the entire monitored area, says Taube. The AI ​​uses this data to calculate a danger value: If it is exceeded, a message is sent to the control center. “An announcement often helps,” says Taube. “You there in the green sweater: Please stop this behavior.” Otherwise, the screens in the control center remained black in order to work as data-efficiently as possible. "Our system is deliberately not capable of facial recognition, so it cannot be misused."

“Technically, an enormous amount is possible”

Due to reporting on violence in public places and the hybrid threat from Russia, social acceptance of video surveillance has increased, says Taube. In addition, technology has developed rapidly in recent years, meaning that the costs of processing large amounts of data through AI have fallen. "Technically, a lot is possible."

A project for AI-supported video surveillance has long been running not only in Berlin, but also in Mannheim, Frankfurt and Hamburg. In Bremen trams, AI is now supposed to recognize and report aggressive behavior. And the Bundestag passed a new federal police law in July that allows automatic real-time facial recognition in certain cases. There are similar plans in many federal states.

The oldest project has been in Mannheim since 2017. There, under the technical management of the Fraunhofer Institute for Optronics, System Technology and Image Evaluation (IOSB), 70 cameras are installed, the images of which are evaluated by software with artificial intelligence. To do this, an AI first recognizes in real time where people actually appear in the images. The next AI converts these image sections into a kind of stick figure in order to then check their movement sequences for possible physical attacks. The system then sends a message to the police control center, which decides whether to send out a patrol. Fraunhofer technology is now also being used in Hamburg. “A stick figure has no identity, no gender, no origin, no skin color, no religious affiliation,” says Markus Müller from Fraunhofer-IOSB in Karlsruhe.

Facial recognition in Frankfurt's train station district

In Frankfurt's notorious Bahnhofsviertel the police go even further. AI-supported facial recognition has even been used there since last year. The technology is intended to compare faces with a police wanted database and thus initially help in the search for missing people and terrorist threats. The AI ​​scans around 300,000 faces every day. Those responsible emphasize that this data will be deleted immediately. A court order is required for any search.

Despite the increasing number of pilot projects, the implementation of the technology in Germany is still in its early stages compared to other countries, says Jan Taube from Staige. Surveillance technologies are commonplace in countries such as the USA and China. In the USA, of course, they are not without controversy. AI-supported camera systems that record passing cars, so-called Automated License Plate Readers (ALPRs), are currently being discussed particularly controversially.

A real resistance movement has formed in recent months against Flock Safety, the best-known provider of such technologies. There is a separate website called “DeFlock”, whose operators are dedicated to the fight against the cameras. Recently there have also been a number of reports of vandals destroying such cameras.

Dispute over flock cameras in America

The flock cameras are about the size of a shoebox, and they have become a familiar part of the streetscape in many areas of America. According to their manufacturer, they can be used to analyze a whole range of information about cars, from license plates, model types and colors to any dents and stickers. The Flock databases can then be searched by customers according to such criteria. According to “DeFlock” surveys, more than 130,000 flock cameras are installed in the USA. The manufacturer itself says it works with more than 5,000 police authorities. Flock claims more than 2,800 crimes are "solved" every week thanks to its cameras, and last year more than 10,000 missing people were found with their help.

Critics question the usefulness of the cameras and see them as a tool for mass surveillance. A few weeks ago, a report in the Washington Post newspaper about almost 50 cases in which police officers were accused of improperly using the Flock systems, for example to spy on their wives or former partners, also caused outrage.

"It's very easy to understand where someone lives and works"

One of the most prominent opponents of flock systems is the Electronic Frontier Foundation (EFF), which advocates for data protection and freedom of expression. For example, it is pursuing a lawsuit against the city of San Jose, which, according to the lawsuit, has almost 500 flock cameras. Saira Hussain, a lawyer for the organization, says the flock systems are much more worrisome than regular security cameras in front of a store. Because there are so many of them and they are all networked, it becomes possible to track a large part of people's movements. “It’s very easy to understand where someone lives and works, or who they hang out with.”

According to Hussain, the immigration agency ICE, which has recently been repeatedly noticed for its brutal treatment of immigrants and demonstrators, can also gain access to Flock data. Flock emphasizes that he does not work with ICE himself, which Hussain does not doubt. But the authority can obtain data indirectly, for example through inquiries to local police authorities.

Amid mounting public outrage, a number of American cities have recently terminated contracts with Flock or disabled their cameras. Even states like Texas and Florida, which are governed by the Republican Party, restrict the use of the cameras. A few weeks ago, Flock was forced to announce a series of changes. For example, the company promises to take stronger action against misuse of the cameras and to delete the collected data more quickly by default. In addition, every search in the database must in future be linked to a specific case and therefore have a specific reason. EFF lawyer Hussain believes these concessions are insufficient. For example, she thinks there should be judicial approval for every search.

The use of surveillance technologies has also been discussed in the US this year in connection with ICE operations in cities such as Minneapolis. ICE officers have often been seen holding smartphones at protesters or cars, presumably to identify them. Among the technologies used by ICE is the Mobile Fortify smartphone app, which allows people to be identified using facial recognition or contactless fingerprinting. To do this, government officials simply need to take a photo of a person using their cell phone camera, which can then be compared with various databases. Another ICE partner is BI2 Technologies, a provider of a system that can identify people within seconds by scanning their irises.

China monitors even remote cities

Surveillance using artificial intelligence has been widespread in China for years. Surveillance cameras are omnipresent, millions of devices leave no corner of public space unguarded in big cities. Countless citizens, including some foreigners, have witnessed police stations using surveillance cameras to easily track individual citizens' routes throughout the city. In some regions, such as Shanghai, the authorities have already presented the systems to the public. Facial recognition for entry control in train stations or trade fairs is part of everyday life.

How widespread the system is is shown, for example, by a tender from the police department in Wuhai in Inner Mongolia in northern China, which the F.A.Z. analyzed. According to official figures, just over half a million people live in Wuhai in an area twice the size of Berlin. At 5:09 p.m. on March 23rd this year, the police department published the result of a tender for the project “Application system for a large-scale multimodal model for urban video and image analysis”. The document, viewable on a Chinese government platform, provides a detailed look at the capabilities that even remote cities like Wuhai find necessary.

The China Mobile Communications Group, which won the contract, undertakes to supply “facial analysis software” and a “facial search matching service”. The package also includes the modules “One file per person” and “One file per vehicle” as well as modules “for merging and visualizing movement histories”, for “relationship analysis” and “for people of particular interest”. A statement from the police department says: “Using the facial recognition system, data sets about people are merged and archived to build up personal files.” The bullet point “Software for connection to the national video standard” suggests that the local systems are networked with nationwide solutions.

All of this is displayed, among other things, on an “all-in-one device for bundling personal image data,” which the police authority has ordered three times. “Within three months” it will also receive a “police knowledge assistant based on a large-scale language model” and an “application service” for the AI ​​provider Deepseek, both of which are used for “urban video and image analysis”. The all-round package also includes algorithms for detecting fires, waste containers, damaged green spaces and road damage.

Can surveillance even prevent crimes?

The descriptions of the tenders are similar to a platform that some international media reported on in the spring. Foreign journalists and diplomats, for example, could be tracked on the platform, which was accessible through an incorrect setting. It was a demo version for the Zhangjiakou Public Security Bureau, where part of the 2022 Winter Olympics took place. The system could access not only cameras, but also train and seat numbers, facial recognition cameras at a ski resort or frequent shopping locations.

“Large language models” are “a consistent further development” for police work, according to a statement from the Wuhai police authority. Artificial intelligence is already deeply integrated into police work today. Now police work should gradually develop from “subsequent collection of evidence and combating ongoing incidents towards prevention”. They want to build an “intelligent system that enables multimodal understanding, early warning and situation assessment” and supports decision-making. According to the document, the police authority charged the package 10,086,000 renminbi, the equivalent of almost 1.3 million euros.

There are many similar projects by Chinese local governments. In mid-May, the police office of the city of Shenyang in northeast China published a tender for the “procurement of intelligent surveillance technology” for almost eight million RMB, around one million euros. This includes a total of around 850 “surveillance cameras for facial recognition”, requirement: at least four megapixels, or almost 600 microphones. Delivery, installation and acceptance must take place within 20 days of signing the contract. A company came into play that took on the contract for around 900,000 euros (just over seven million renminbi) and is now primarily installing cameras from the Chinese Hikvision group.

Many things are not legally possible in the EU

Much of this would not be legally possible in Europe. The EU AI regulation prohibits