Facial recognition in Frankfurt am Main | “Real-time detection is a serious encroachment on fundamental rights”

The police praise facial recognition in the train station district of Frankfurt am Main; there is said to have already been a hit. The system is now to be installed in other locations. Photo: dpa/Boris Roessler The Hessian police are testing facial recognition in the train station district of Frankfurt am Main; it is the only such system in Germany to date. Is that why you're suing?

Yes, that's exactly why. Hesse is the first federal state to create a legal basis for biometric real-time remote identification to avert danger. Frankfurt am Main is the first city to use such an AI system with cameras in public places since July 2025. If it is hit, it reports to the police, who can take follow-up action.

There is also AI video surveillance in Mannheim, Hamburg and Berlin. What's the difference?

In Frankfurt, the system records facial features in real time and compares them, for example, with a mugshot. In Mannheim, Hamburg and Berlin it is about recognizing behavioral patterns that indicate crimes.

The Federal Criminal Police Office has also had a facial recognition system since 2008 that the federal states can use.

The BKA system subsequently compares photographs taken with photos from the Inpol database, the police information system. The main purpose is to investigate crimes or check the identity of asylum seekers. In Frankfurt, however, facial recognition is enabled in real time and linked to cameras in public places. It serves to avert danger.

This is now to be expanded, also geographically.

According to the Hessian Ministry of the Interior and the police, the pilot project has been successful so far. The expansion to Konstablerwache and Hauptwache has been announced. It can be assumed that it can then be used across the board in Hesse. In the Federal Police Act, the Bundestag and Bundesrat have also decided that real-time remote identification is also permitted for the federal police; the law came into force on Wednesday. As far as I know, Saxony has already adopted a legal basis. Baden-Württemberg and Schleswig-Holstein have presented or announced draft laws.

Davy Wang is a fully qualified lawyer and process coordinator at the Society for Civil Rights (GFF). He works on digital freedoms and is primarily concerned with state surveillance, data collection, and AI and platform regulation.

Frankfurt police say they are looking for terror suspects, missing people and victims of human trafficking. Have any been found yet?

In the fall of 2025, there was a media report praising that a missing 16-year-old had been found. I haven't heard of any other cases.

We do not know the error rates. We also don’t have the training data. The AI ​​system's decision-making logic is incomprehensible to neither the officials nor those affected, and there is no public information about it. We know that there is a minimum similarity threshold that the police are allowed to set for each operation. If you lower it, many more people will be captured, even those who only look similar to the person you are looking for. If you set it very high, there may be no hits at all.

If the minimum similarity threshold is set low, are there more follow-up measures against uninvolved people?

The police say that in the end it is not the software that decides on measures, so it is not an automated system.

It is an automated AI system that independently compares faces and reports hits. But there is the problem of “automation bias”: people tend to trust machine suggestions more than their own judgment. There is also little time to check the result. In addition, it is not clear to the officials how the decision-making process works. If there are risks of discrimination that the police are not aware of, it is difficult for them to ensure that these do not materialize.

They are suing together with the Frankfurt advice center Doña Carmen, whose entrance has a camera pointed at it.

We are suing two board members. The camera location and the use of technology limit the functionality of the advice center. This can have a major deterrent and intimidating effect on sex workers who rely on low-threshold and confidential advice. If they have to worry about being regularly biometrically recorded on the way there, they may no longer come even in acute emergency situations.

We also want to submit a submission to the Federal Constitutional Court. It is intended to review the legal basis for AI video surveillance and set standards and limits on this very new power. The current legal basis is intended to partially declare it unconstitutional.

What about political meetings? In Berlin the cameras then have to be switched off.

From what I understand, systems in Frankfurt are covered with blinds or turned around so that they do not record meetings. If people have to walk through camera fields on the way to or from the meeting, they can still be recorded. It is also questionable whether the police can react quickly enough to cover the cameras in time during a spontaneous emergency meeting.

Which other groups of people would be affected by intimidation in the Frankfurt train station district?

Residents, people who work there and people who receive services such as psychotherapy there are particularly affected. This also applies to groups that are already stigmatized, such as people with addictions or homeless people who are dependent on assistance facilities such as homeless assistance, social counseling or medical care.

You mentioned training data. Videos of the operations are also sent to the system manufacturers.

In Berlin, the renewed police law contains a blanket authorization according to which the police can use all personal data from their databases for the development and training of any systems. Clauses state that personal data must be anonymized. If this is only possible with disproportionate effort, clear data may also be used. This formulation is too vague. This is particularly relevant if this data goes to private providers.

In Berlin, behavior recognition is based on high-resolution cameras. Would switching to facial recognition be a big step?

Technically they are two different systems. If a legal basis for facial recognition were created in Berlin, the software could be connected to the cameras.

In your opinion, what should a constitutional law for facial recognition look like?

Facial recognition using cameras in real time is a particularly serious encroachment on fundamental rights, and its use must therefore be severely restricted. It is actually only permissible to protect particularly important legal interests such as the life, limb or freedom of people. There must be actual evidence that this is imminent. The minimum similarity threshold must be determined by law. The law must regulate appropriate storage periods, as well as what happens to data that is marked as a non-match. There must be deletion obligations. And if a hit occurs, there must be an obligation to notify those affected.

Then would you give the green light?

Since it is a very serious encroachment on fundamental rights, the state should not use real-time facial recognition systems. If so, then only under the conditions mentioned.

What is the status of your lawsuit in Frankfurt?

We filed the lawsuit in December 2025 and the detailed statement of claim in April 2026. We expect a hearing to take place sometime next year.

The GFF is also working on other surveillance technologies, in particular biometric comparison with internet data. What's new?

This has a completely different dimension and also affects highly private images behind access barriers or paywalls, for example on social media. Or images that were not uploaded by those affected themselves and that we don't even know about. In order to carry out the comparison in a technically practical manner, the police would also have to create a facial database themselves. This would be inadmissible under the EU-wide AI regulation.

The draft laws for the BKA Act and the Federal Police Act contain a clause that would allow the police to use private providers. Providers such as Pim Eyes and Clearview AI have created such facial databases. These providers collect the images in violation of EU law and the General Data Protection Regulation. If the police are allowed to resort to this, they are using illegal offers to circumvent their own legal obligations.

Would you also complain against this?

During the legislative process, we have already criticized the draft on constitutional and Union law grounds. If the law is passed, we will continue to work on it.

The GFF is also working on forensic software for reading cell phones that police and immigration authorities use.

The police are allowed to do this if there is an initial suspicion of a crime and the device would be relevant for evidence. However, because of the many types of data on a cell phone, this can always be the case. The immigration authorities act on the basis of the Residence Act: If the identity or nationality of a tolerated person is unknown and they therefore cannot be deported, cell phones and other data carriers may be confiscated. What is critical is that the immigration authorities are now allowed to keep cell phones until they leave the country.

According to the justification for the law, the regulation is intended to prevent people who are required to leave the country from thwarting or delaying their exit and deportation by destroying identity-relevant documents. I don't understand why a seizure would still be necessary after the data has been read.

Police say they lacked forensic capacity, so cell phones were left lying around longer in low-level cases. Isn't that also a serious encroachment on fundamental rights?

In any case, this affects the fundamental right to property. Nowadays everyone relies on their cell phones for everyday life. This is also relevant for the immigration authorities, because they are dealing with tolerated people who often receive social benefits and are certainly not in a position to buy an equivalent replacement device.

The GFF is also looking at Palantir, a kind of “system of systems” that ties together data from the measures we have talked about here.

This is definitely a danger to fundamental rights, especially if an AI system creates an evaluation that you cannot understand yourself. With regard to the Palantir application in Hesse, the GFF has fought for a decision from the Federal Constitutional Court, which has set out very specific requirements for a legal basis. Concrete protective and procedural precautions are needed to ensure that discriminatory algorithms do not emerge. And if they do exist, they must be recognized and corrected.

Then you wouldn't have any problems with it in the end? I ask this because it is a dilemma to go to the level of law: If there is no suitable law, it will be made and the measure will be applied anyway. In the end, with your blessing, a comprehensive monitoring instrument could be used.

I wouldn't put it that way. We are a legal NGO. What is relevant to us is what could be legally challenged and what minimum constitutional requirements there should be for the use of such technologies. Of course, politicians can decide not to push these limits, not to pass such laws and measures, or to impose even stricter requirements. However, with our strategic litigation we can only argue constitutionally.

With us, no one has to pay to be informed. Sounds like a bad business model? It is too. But independent journalism and participation for everyone are more important to us than profit. Journalism still costs money: research, printing and all the people who do it. That's why we depend on you. We need 300,000 euros by the end of the year.

Donate once or regularly, every amount counts. So that nd remains.