Police use of AI for real-time facial recognition

Adam Marttinen (SD)

Mr. Speaker! We are debating the Committee on Justice's report on the Police's use of AI for real-time facial recognition. In summary, the committee supports the government's proposal for a new law which, among other things, means that the Police Authority will be given the opportunity to use AI systems for real-time facial recognition in order to be able to combat crime.

The new law means that the Police Authority shall be allowed to use AI technology if it is absolutely necessary to locate or identify a certain person who, for example, is suspected of being a victim of a crime such as murder, human trafficking or human exploitation, or who is missing and is suspected of being a victim of a crime. It can also involve a situation where there is an imminent risk that a person will commit a serious crime involving danger to another's life or physical safety, or that a person can reasonably be suspected of having committed a certain serious crime for which the penalty scale includes imprisonment for four years or more, or that such a penalty can follow for attempted, prepared, or incited such crimes, for the purpose of investigating or prosecuting the crime. It can also involve persons who have been convicted of such crimes, where the purpose is to be able to execute the criminal penalty.

The use of the technology shall be proportionate, and a permit from a prosecutor or a court shall be required. In urgent cases, however, the use may be commenced without a permit, and in such cases, an application for a permit shall be made within 24 hours.

This is a legislative proposal that the Police Authority itself has requested, and which the government and the Sweden Democrats consider to be well-balanced in light of society's need to have an effective crime-fighting authority that puts people's lives and safety first. One can take imminent terrorist attacks as an example and how important it is then to be able to identify suspected perpetrators within seconds who, if they had been at large, could cause very great damage to society and to individual individuals.

In normal cases, it might take weeks to identify a perpetrator who has committed a serious crime. Now, instead, it is a matter of seconds. This is a type of efficiency that I believe the people at large will support. Against the background that there must be a level of suspicion, this is not about any major infringements on personal integrity. It must concern serious crime, and it also requires a decision from a court or a prosecutor in this case.

This is, therefore, an example of what the Sweden Democrats and the government are moving forward with. We can also see in the report the division in the opposition, which is not actually unique. The division that we can see in this report, where, for example, the Centre Party is against it and the Left Party and the Green Party have objections and limitations and suggest that it should not be as effective, we see report after report.

I am very happy and positive that the Sweden Democrats and the government have been able to move the positions for law enforcement agencies during this mandate period and achieved actual results out in reality, where we see that explosions and shootings decrease with each passing year and serious criminals are prosecuted even if they are located abroad.

We are therefore moving the positions in many ways and making it possible to push back crime. At the same time, we remember what eight years of a red-green government means, namely grit in the machinery of the justice system's opportunities to combat serious crime. We can see that in this report, just as in many other reports when it comes to the Justice Committee's work against serious crime. I think that is something the voters should take away.

We have a government that prioritizes effective law enforcement agencies while we consider the rule of law and privacy and succeed in reaching an agreement on criminal policy proposals. On the other side, they are divided, and that is what awaits if, God forbid, there is a change of government this autumn: a divided opposition that will argue more than deliver in the area of the justice system.

This is for me finally, Mr. Speaker, to motion for approval of the committee's proposal for a decision and rejection of the opposition's various motions.

The speech at riksdagen.se, in Swedish (opens in a new tab)

Jan Riise (MP)

Mr. Speaker! I could not let the opportunity pass without offering a brief rebuttal to Member Marttinen's claims.

I naturally took particular note of all the discussion regarding the opposition's split in this context. I can only say that I do not believe that Adam Marttinen needs to be so worried about it. We will probably resolve this. I can see great advantages in having four parties that have slightly different things; then the voters can decide a bit in which specific direction we should go.

When it comes to the proposal itself, there are questions that the member perhaps went over a bit too quickly and which deserve to be discussed a bit more closely. I want to recall that recognition in a few seconds was mentioned on a couple of occasions. This means that a large number of people are to be identified in a few seconds, and then one is to find the individual person. This is based, among other things, on access to enormously large image databases. We say that it is forbidden to use anything other than Swedish databases – the passport police, the police naturally, visa authorities and such. However, it has occurred, which in itself is forbidden, that one has used American databases where the internet is scraped for images that we have posted on Facebook and other places.

I can see some clouds of concern in that: partly which databases we should have, and partly how you, in your reasoning, handle the issue of the false positives that inevitably arise. That there is 99 percent accuracy in the lab is one thing, but in reality, it does not. It might be much lower, and we must be able to handle that in a good way.

The speech at riksdagen.se, in Swedish (opens in a new tab)

Adam Marttinen (SD)

Mr. Speaker! When it comes to the divided opposition, I will let the one who examines our reports decide whether an image of unity arises. I think that the reservations in this report constitute an example of precisely that division, where one of the parties that is to be part of the government base rejects the proposal in its entirety and the other two parties, in addition to the Social Democrats, have objections when it comes to proportionality, the degree of suspicion, and limitations of the legislation in other ways.

I think it is absolutely reasonable to have a debate about privacy and how the identity of identifiable persons should be handled by our authorities. It is a fully legitimate debate. But as I explained, and as the bill in particular explains, there are well-balanced boundaries.

One can discuss AI broadly and extensively, but it is also a guarantee that the human eye does not need to process the identity of different people who may appear in a scan, as it is done by a function that does not violate personal privacy. There are therefore both advantages and disadvantages with the progress of the technology.

We believe that the limitation we have made is well-balanced. In this situation, the technology creates such efficiency gains for the authorities that we believe it is necessary to keep pace with the population's perception of how efficient the authorities should be when it comes to combating crime.

The speech at riksdagen.se, in Swedish (opens in a new tab)

Jan Riise (MP)

Mr. Speaker! Thank you, Adam Marttinen, for the answer!

I can return a little bit to who should decide. I can agree that the one who reads our reports will naturally have to make their own assessments. In the end, it is the voters who decide, and then it is probably good if there are variations to consider when it comes to who or whom one thinks should get that precious vote.

There are some different views regarding privacy. I noted that the member spoke of the privacy intrusion as quite limited for those persons – the Speaker, the member, I and others – who will be identified in various contexts. One can naturally argue that we – I hope – have clean hands and thus nothing we need to worry about.

Research nevertheless shows that there is some risk that one changes their behavior if one knows that one is being observed everywhere. One might avoid places. One might particularly do so if one is there in some company that for its part is not forbidden but which can have other implications. AI can, after all, put together a lot of things. "Oh, Member Riise has been seen there, there, and there over a sequence of months or weeks. How should we interpret that?" And then it can become known if it spreads.

I would like to ask how the member views the privacy issues in a broader perspective than that we have to trust that those who are not relevant are filtered out. I previously mentioned that false positives might be subjected to some form of intervention. But in broad strokes – will this change our reality in any way?

The speech at riksdagen.se, in Swedish (opens in a new tab)

Adam Marttinen (SD)

Mr. Speaker! I think this is an interesting philosophical debate. I participate in it in many ways beyond the debates held in the Riksdag. I think it is well worth reflecting on such things. What we can take away is that the popular acceptance for, for example, camera surveillance has been much greater than the political acceptance has been through the decades.

Cameras have taken up a larger space in our societies. One can philosophize about how the public perceives that there is a camera. Does the fact that one is being monitored automatically mean that one avoids an area or avoids certain behaviors? I do not think that is the case.

Given the limitations in the bill, I do not believe that cameras in the city will cause the population to limit their movement patterns in any way. It would likely be the opposite. In the bill, one can read that the use of this shall be limited to persons who are suspected of serious crimes. I believe, therefore, the opposite: that people may move more in, for example, city-center environments rather than avoiding them for fear of serious crime.

People's freedom from crime is a perspective that weighs very heavily in this debate. If one risks being exposed to serious crime or other violations, one avoids public environments. This can be avoided through a certain type of surveillance that provides some form of security for those who are in city-center environments and elsewhere. I think we have a good balance in this, while the split in the opposition is clear, as I stated in my first contribution.

The speech at riksdagen.se, in Swedish (opens in a new tab)

Gudrun Nordborg (V)

Mr. Speaker! I can begin by saying that Vänsterpartiet, on the whole, supports the committee report, but that we have two important reservations together with Miljöpartiet. I move for approval of reservation 2. I will return to that.

I would also like to comment on what Adam Marttinen just said about us in the opposition being in disagreement and that the Left Party and the Green Party had objections and wanted to make limitations. I would like to assert that our positions instead concern strengthening the rule of law aspects of the project with facial recognition.

It is, as was just said, important to see that it is in very serious situations that this may come to be used. It should be about seeking a specific person, i.e., not mass surveillance. There must be serious suspicions and an imminent risk that the person commits new serious crimes.

It is also the case that one will attempt to track those who may be victims of crime in serious situations such as human trafficking, human robbery and human exploitation. It is therefore not something that will be frequent. These crimes do not occur constantly and always or everywhere, even though they obviously occur. It is therefore why it can be important to have the possibility of facial recognition.

In our reservation, we point out, among other things, that we believe it should always be a court that makes the final decision and that it should not be sufficient for a prosecutor to do so. The reason for this is that we believe the court is safer. Someone may object that it then takes longer. Perhaps that is the case, but if a delay poses a danger and the matter must be handled quickly, there is also an opportunity for the police to act and seek permission within 24 hours. That should be solvable in a good way.

We also believe that legal certainty should be strengthened by having a public representative and the right to appeal through the representative. Notifications should occur in any case where facial recognition has been used.

A crucial question is, of course, where the information goes. Does it stay with the police? That, we mean, is an inescapable requirement. In the police's oversight council, I have asked how it works. We have also summoned the National Police Commissioner to the Justice Committee to ask questions. A concern has concerned a company called Palantir. The National Police Commissioner and the staff refuse to disclose which companies or platforms they use. But they say they can guarantee that the information they collect stays in Sweden, with the police. It is not passed further. This must, naturally, be followed up.

We also want there to be requirements for ongoing testing of how this is used. One will be obliged to report the use to the Swedish Authority for Privacy Protection. It will be a continuous control. One must report every time facial recognition has been utilized.

It is also, I believe, important to say that one is not collecting new information. One uses the camera information that exists through the surveillance we have permitted so far. It concerns specific, pre-named or identified persons. The Swedish Authority for Privacy Protection shall have full control over this.

One must also consider what the alternative is. The police already have certain possibilities today to use automatic image analysis in criminal investigations. This occurred by the Data Inspectorate granting permission for this in 2020. The police are therefore already using automatic image analysis today, which can involve the handling and processing of biometric data.

The fact that the police are already using the technology in criminal investigations means that we in Vänsterpartiet have chosen to accept the government's proposal, also for the reason that we think it is irresponsible to leave such a sensitive area without clear rules for the time being. It does not appear to be an alternative.

The proposal also stays within the framework of the EU's legal act on artificial intelligence, the AI Act, which already applies to us in Sweden. Vänsterpartiet voted in the European Parliament for the regulation because it entails necessary regulation of the use of AI across the entire EU.

A friend of order might think: What does the regulation look like? It contains, in itself, a principled prohibition against using AI in systems for real-time facial recognition. But it is important to mention that the rule makes it impossible to use such mass surveillance as occurs in certain authoritarian countries.

A member state may, however, despite the principled prohibition, decide in its national law to prescribe possibilities to allow the AI system for real-time facial recognition in publicly accessible places for law enforcement purposes. This is what we are about to introduce today. We stay within the framework, we do it with legal certainty guarantees, and we would very much prefer to do it with additional legal certainty guarantees, something that Vänsterpartiet and Miljöpartiet have proposed. I hope, therefore, that you vote for reservation 2.

The speech at riksdagen.se, in Swedish (opens in a new tab)

Ludvig Ceimertz (M)

Mr. Speaker! For many years, Sweden had a criminal justice policy that did not keep pace with developments. At the same time as serious organized crime became more brutal, more international and more technologically advanced and grew ever stronger, the justice system was prevented from working effectively and was forced to use outdated tools. This government is working to change this step by step.

The deadly gang violence has been halved in recent years. The clearance of the most serious violent crimes has increased significantly. For example, the clearance rate for shootings has gone from approximately 25 percent to perpetrators being arrested in almost all cases in 2025. Furthermore, more gang leaders are being arrested and prosecuted. This is a result of a consistent realignment of Swedish crime policy, and it is delivering results.

Mr. Speaker! Naturally, a broad palette of measures is required to push back crime. No single reform is sufficient on its own; instead, a consistent direction is needed for a safer Sweden. This course must remain firm.

Mr. Speaker! An important part in combating gross violence has been the expansion of camera surveillance in Sweden. The Moderates were clear during the 2022 election campaign about wanting to increase the surveillance cameras in Sweden. A goal was set of 5,000 police cameras compared to the 948 cameras that existed then. Since the Moderate-led government took office, the police's camera capacity has been significantly expanded. At the end of 2025, the number was almost 4,000. We are close to the goal of 5,000.

Mr. Speaker! The Moderates want to continue expanding the camera surveillance and have now set a new goal of 10,000 cameras by 2029. It is about more fixed cameras, more mobile cameras, more drones, and more ANPR cameras, which can read registration numbers and help the police track criminals.

This has not happened by itself, Mr. Speaker. It is a result of political decisions. The Moderate-led government has implemented extensive reforms to make it easier for the police to use camera surveillance. The permit processes have been simplified. The bureaucracy has been reduced. Furthermore, the police have been given significantly greater opportunities to quickly set up cameras in crime-prone areas and in places where serious crime occurs.

The results are clear. The camera helps the police identify perpetrators. It helps to secure evidence and solve murders, explosions, and shootings. Camera surveillance is today a completely central tool for preventing and investigating crimes and in this way increasing safety in Sweden. All people should be able to feel safe in their residential area, on the way home from work, and when children go to school or practice.

The development should not stop there, and here this bill on real-time facial recognition comes in in an important way.

Mr. Speaker! Criminal networks use modern technology every day. They have encrypted phones. They move quickly between different locations and plan violent crimes digitally. They try to avoid detection by constantly staying one step ahead. Therefore, the police must also have modern tools. That is exactly what this bill is about.

The police can already today use facial recognition retrospectively in criminal investigations. If a surveillance camera captures a suspected perpetrator, the police can analyze the material retrospectively and compare images against registers or other material, even with the help of AI.

What is now being proposed, however, is that the Swedish Police Authority and the Security Service also get the opportunity to use AI-based facial recognition in real time. The difference is important. It is no longer just about being able to solve crimes after the fact, but about intervening while something is happening, or even better, before it happens. It is about preventing serious crime where crime victims have their lives destroyed.

Mr. Speaker! Let me give an example. Imagine that a child has been abducted. The police have access to pictures of the child and suspect that the child is being transported through large public environments, such as train stations, shopping centers, or airports. Today, it is fully possible for the police to capture movement patterns on cameras, but they are not allowed to use AI to find or sort through it.

With this legislation, the police are given the opportunity to use AI-based facial recognition in real time to quickly locate the child and intervene before the situation worsens and before a child is taken abroad. It is therefore not about technology for technology's sake, but about giving the police the opportunity to save lives and protect people in situations where every minute can be decisive. For this very reason, many other countries already use such technology.

Mr. Speaker! This is not a system for general surveillance of ordinary people. It is about using the technology targeted at specific individuals in cases of very serious crime. It is also important to understand that the EU AI Act explicitly allows for exactly this.

The starting point in the EU is that real-time biometric identification shall be prohibited. But the EU has simultaneously noted that there are situations where democratic countries must be able to use the technology to protect people and combat serious crime. Therefore, there are already built-in exceptions today for, among other things, terrorism, human trafficking, human smuggling and other serious violent crime. It is these exceptions that we are now using.

Mr. Speaker! There are those who argue that the state should not use this type of technology. I do not share that view. The alternative is not a society without technology. The alternative is a society where the criminals use the technology while the rule of law abstains, and that would be irresponsible.

When the police have the opportunity to prevent an act of violence, find a kidnapped child, or arrest a dangerous perpetrator before more people are harmed, I believe that society should also give the police the tools to do so. Ultimately, it is about safety and about people's freedom to live normal lives without fear of crime and violence.

Mr. Speaker! This government is building up a modern and strong crime fighting step by step. We ensure that the police get more cameras and better opportunities for camera surveillance and that the police can use the technology more effectively against serious crime.

The Moderates want to continue on the path taken and stand firm on the goal of 10,000 police cameras in Sweden. We know that cameras, used correctly, help the police to prevent crime, prosecute perpetrators and create safety for people. With this, I move for approval of the bill.

The speech at riksdagen.se, in Swedish (opens in a new tab)

Ulrika Liljeberg (C)

Mr. Speaker! I drove yesterday from Leksand to Stockholm for this debate and this week's parliamentary work. It was a fine summer day in our beautiful Sweden. I listened to Sveriges Radios podcast Brottsutredarna about thefts of ATVs, motocross bikes, enduro bikes and tools.

A guy from Leksand whom I know, Erik, told me about how last summer, after a burglary in his village in Leksand, he called the Swedish police with decisive evidence in real-time regarding where the stolen goods were located. He was given some kind of variant of "don't call us, we'll call you." Erik instead got in contact with a police officer in Poland, where the stolen goods were, which led to the discovery of a large amount of stolen goods. This happened before the Swedish police had even called back, let alone arrived on site. This was the Tidö government's police in 2025.

It should be said that the police in Dalarna have acted since the estate was found and that a trial is ongoing in Falu District Court.

In the podcast, it was also stated that one of those involved had been arrested by the police repeatedly with stolen goods but had always been released. This was in 2025.

What I want to say with this, when today's debate is about facial recognition using AI in public places, is that the police's work is to prevent and deter. If a crime has occurred, it should be investigated and prosecuted. Here we see that the police, despite the enormous increase in resources, are not solving more crimes. That which was stated in the podcast—being arrested for theft several times and released every time—does not have any direct deterrent effect.

My objection to today's proposal is that in order to introduce tools that are intrusive and monitoring, one needs to have fully utilized the tools already available. The police's basic work needs to be significantly improved before introducing tools that challenge personal freedom in the way today's proposal does.

It must be said that the Center Party has been in favor of most of what has come from the Tidö government. It has been necessary and important. We are putting the crime victim in a better focus than we ever have before. But we have said no to things where there have been legal certainty risks and where the tools are not as sharp as they are presented.

Visitation zones were introduced despite strong criticism. They have been used six times in recent years: four times in 2024, once in 2025, and once so far this year, of which the latest was on the grounds of the conflict in the Middle East.

Regarding anonymous witnesses, there is great concern for legal certainty and that it would make valuable witnesses quite worthless. It is now being said in SVT Nyheter that one may now, after having had the opportunity to use anonymous witnesses for nearly 1.5 years, potentially receive such a case.

These are not the sharp tools that are being portrayed, and one must be careful with tools that challenge personal integrity.

I want to say it once again: the Centre Party will never accept a crime development like the one that has occurred in recent years. The gangs' brutality, the shootings, and the insecurity that is spreading require a greater focus on the crime victims, tougher measures, and effective preventive actions. We are prepared to continue providing the police with the tools they need.

Mr. Speaker! Not every tool is the right tool. Not every measure is a good measure. This proposal is not about being for or against new technology. This is about where one draws the line between crime fighting and surveillance and about how purposeful and effective each proposal is.

The starting point in the EU's so-called AI Act is that it shall be prohibited to use systems for real-time biometric remote identification of physical persons in publicly accessible places for law enforcement purposes. This is an important starting point in these debates.

As reasons for this, it is stated, among other things, that such use particularly encroaches on the rights and freedoms of the persons concerned. The use can affect the private lives of a large part of the population, create a sense of constant surveillance, and indirectly deter from the exercise of freedom of assembly and other fundamental rights. It is also specifically stated that the fact that the controls occur in real-time means that the effects of them are immediate and that there are limited opportunities for further control or correction. It also highlights the technical deficiencies in certain AI systems.

It is about introducing a technology for mass surveillance that risks undermining the very foundations of our free and open society. Therefore, the Center Party says no to this proposal. I am not saying that it is easy considerations every time. But unlike, for example, the Tidö parties and the Social Democrats, we also say no to chat control. There is an important purpose, but we make a different assessment of the whole.

The Center Party is not anti-technology – on the contrary. We believe that AI, when used in the right way, can be a fantastic tool to strengthen societal development and streamline the police's work. But it is important that it is done in a legally secure and responsible manner, and we do not believe that it has been done.

Those who have been subjected to human trafficking, human robbery, human exploitation, and exploitation for sexual services are one of the most vulnerable groups. If we truly want to prioritize and protect them, the Tidö parties should address this in the regulatory letters. There is nothing about human trafficking in the regulatory letters. On the other hand, it mentions the 5,000 cameras that have been mentioned here, and we stand behind those, but one must also have a will and a direction when it comes to what one should be looking for. The regulatory letters need to be supplemented with precisely human robbery.

We know that black labor is now the single largest source of income for serious organized crime. Black labor is present on construction sites, in restaurants, and in service occupations. We need to do more for these people. Do not use them as reasons to monitor in public places when we know that they work almost around the clock, sleep on mattresses on the floor, and are not in public places.

Organized crime poses serious threats to individuals but also to society as a whole. Therefore, do what is effective and do not use these people as a reason for mass surveillance.

Fighting crime is to defend the open society. But we can never defend the open society with methods that abolish it. The Centre Party chooses another path – a path that combines powerful crime fighting with an unwavering defense of the rule of law and the individual's civil and political rights.

With that, I move for approval of the Centerpartiet's reservation 1 in the report.

(Applause)

The speech at riksdagen.se, in Swedish (opens in a new tab)

Martin Melin (L)

Mr. Speaker! I do not believe there is anyone in here who knows who Frans Oscar Roos was. I say was, for he is no longer with us. Frans Oscar Roos was the first Swede who could be identified and prosecuted after committing a crime through his fingerprint.

The reason it was possible to do so was that, at the Stockholm Police's identification office, which was then located within the detective department, only a year before he was arrested, a register of criminals' fingerprints had been established. Frans Oscar Roos was thus one of all the criminals who, the year before, had been required to leave their fingerprints with the police.

This was new technology. Being able to dust a crime scene and find fingerprints and then compare them with a register of known criminals was new technology, which gave good results. It did, however, take a couple of years before the district courts considered it to be evidence that held up and could be introduced in court.

It was 120 years ago this year. I promise you that back then, just like now, there were those who thought that this thing with a register was unpleasant and wrong and that it was an infringement of privacy to register people's fingerprints.

Mr. Speaker! Today we are debating new technology: the police's use of AI for real-time facial recognition. And there are now, just as there were 120 years ago, those who think it is disgusting and an infringement of privacy.

It is a self-evident matter for me that the police and prosecutors should be given access to new, effective technology to be able to prevent and investigate crimes and arrest dangerous individuals who have committed serious crimes. We are talking about murderers, terrorists and rapists. But it is also about being able to find crime victims, for example those who have been exploited in human trafficking. We actually would not need to stand here and debate this today.

If the police are to search for a dangerous person, there are two ways to do it. The police can sit in their police cars with a photograph and compare it with all the people they see on streets and squares. One can also ensure that there are cameras in strategic places all around society that alert the police when they recognize the person who is wanted. It is old technology versus new technology.

Finally, Mr. Speaker: I and the Liberals want the police in Sweden to have the best technology, so that Sweden becomes a safe and secure country.

With this, I move for approval of the committee's proposal.

The speech at riksdagen.se, in Swedish (opens in a new tab)

Jan Riise (MP)

Mr. Speaker! The Government proposes that the police should be granted the right to use real-time facial recognition with the help of artificial intelligence. This is well known here now.

The overriding heading is: "The need for effective tools in crime fighting." The argument is therefore that serious crime is a major problem and that real-time facial recognition is an effective tool, among other tools of course. I actually have no objections to that. But I think the problematization in this bill regarding technology and expected outcomes is a bit meager.

Mr. Speaker! I will therefore give a short version of the history. The technology is around 60 years old. The development gained momentum after the terror attacks in New York in September 2001. As you remember, the focus on surveillance, not least at airports, became an important concept. Access to large databases in combination with funding and improved technology made it all possible.

American films and TV series, such as Minority Report, a film from 2002, and perhaps especially the series Person of Interest, which aired from 2011–2016, are often mentioned as references to this development. I believe the latter is still available on one of the major streaming services.

The technology has, of course, developed significantly and reaches almost perfection in a lab environment, but it is not quite as good in reality – that was what I was trying to point out earlier – when, among other things, lighting conditions and resolution are not perfect. It risks leading to a large number of incorrect identifications and, in the worst case, interventions against innocent people who actually just happen to be in the wrong place at the wrong time.

This requires an increased focus on issues of ethics and privacy, accountability and transparency. It is inherent in the nature of the matter that people who are identified are unaware of this until an intervention occurs. As an individual, one therefore cannot escape, and the entire system is sometimes described as a paradigm shift in surveillance. There are plenty of examples of bias and incorrect identifications. Unfortunately, it has so far affected, not least, people of color and women, peculiarly enough.

Mr. Speaker! Another pressing issue, which is not actually discussed in the bill, is where the images used for recognition come from. In Sweden, it is not permitted to use image databases from, for example, the American Clearview AI, which is the company that vacuums up images from Facebook and other social media and wherever they can be found. It is, as mentioned, not permitted in Sweden, but it has occurred. The Swedish police shall, in principle, use state registers such as driver's licenses, passports, visas, own photos, and possibly also images from surveillance cameras and body-worn cameras.

Mr. Speaker! We also need to mention the EU AI Act. The EU's regulation on artificial intelligence, the so-called AI Act, is already applicable law in Sweden. In the European Parliament, we in Miljöpartiet have stood behind the regulation, because a strong common regulation of artificial intelligence was needed that safeguards safety and fundamental rights.

The starting point in the regulation is a general ban on real-time facial recognition in publicly accessible places. It is an important principle. It sets a clear limit against a development where the state normalizes mass surveillance in the public space. I repeat: the starting point is a general ban. But I realize that in the regulation, just as Member Ceimertz touched upon – in my own words then – there is a limited scope for member states to allow AI under strict conditions for certain law enforcement purposes. It may not be an open door, but it exists there as a possibility.

For us in Miljöpartiet, it was important that the AI Act should be implemented in its entirety, as well as the ban on real-time facial recognition with only a few limiting exceptions. We are not there yet. The last part of the AI Act has not been finalized but has been pushed forward. That is where these high-risk applications are. I think that is very unfortunate. I view with great concern the consequences that the use of AI systems for real-time facial recognition can have for people's private lives, privacy, and civil liberties.

The Institute for Human Rights has warned that the proposal entails a restriction in the protection of private and family life and that it could also affect freedom of expression, freedom of demonstration and assembly, and freedom of religion. Civil Rights Defenders has pointed out the risk of discrimination, racial profiling, an increased sense of control, and a development where people might refrain from being in or expressing themselves in the public space. That type of chilling effect does not belong in an open and democratic society.

Miljöpartiet also shares the concern expressed by several referral bodies that the government does not sufficiently clearly answer central questions about how the legislation will function in practice. The bill provides no clarity on which companies shall deliver the systems, what technology shall be used, how accuracy shall be ensured, how incorrect hits shall be handled, or how the state shall guarantee that data is not used by private suppliers for other purposes.

When it comes to this type of technology, the government's assurance that the systems will be used responsibly is not enough. The requirements for transparency, control, and independent review must be high from the very beginning. We want to remind that this is a technology that poses significant risks, not only in film and TV but also in reality, not least when it comes to discriminatory outcomes and the rule of law.

The Green Party's starting point is therefore that the application of the proposal must be more restrictive than the government proposes in the bill. But we understand that there may be situations where the technology can be justified, for example, to retrospectively examine material within the framework of criminal investigations to find missing persons or victims of human trafficking or sexual exploitation and to avert an imminent threat to life and safety.

It must be ensured that the information generated through the use of AI systems for real-time facial recognition also remains with the police. Data must under no circumstances risk being spread to or used by private suppliers. When advanced analysis and surveillance tools are purchased from private actors, significantly stronger control and protection for information, traceability, and independent oversight are required than the government reports.

In order to ensure that the use of AI systems does not lead to discrimination and mass surveillance, mandatory recurring risk analyses should be carried out. Incorrect identification in a law enforcement context can have very serious consequences for the individual. We therefore need ongoing testing, documentation, review, and follow-up.

It is particularly important that proportionality can be assessed in a good way, not only based on the interests of crime fighting but also based on the technology's impact on privacy and freedom of movement, and with consideration for the fact that people should not feel anxiety that they risk being recognized in the public space.

Finally, Mr. Speaker: In order to ensure that AI systems for real-time facial recognition are applied in a legally secure, proportionate, and lawful manner, the use must be evaluated before the bill can be made permanent. Such an evaluation could, for example, include how often the technology has been used, in which situations, with what results, how often false hits have occurred, how proportionality assessments have been made, whether certain groups have been disproportionately affected, and how the reporting obligation and supervision have functioned. We therefore propose a time limit of, for example, five years.

I naturally stand behind all three of our reservations but move for approval only of reservation 3 under point 3 regarding evaluation and degree of suspicion.

(Applause)

In this speech, Jacob Risberg (MP) agreed.

The deliberation was hereby concluded.

(A decision was to be taken on 26 May.)

The speech at riksdagen.se, in Swedish (opens in a new tab)