Sweden Pointed a Camera at Its Own Values

The Riksdag just voted to let police scan your face in real time — and the argument for it is harder to dismiss than it should be.

I noticed the security cameras before I noticed anything else. Walking through Hötorget on a Tuesday morning last autumn, I found myself counting them — the clean white domes mounted above the entrances, the discreet units bolted to lamp posts, the ones swiveling almost imperceptibly on the roof of the Kulturhuset. I'd walked that square hundreds of times. But something about the mood that week, after another shooting in Skärholmen had made the front pages, made me look up. Sweden's cameras had been there all along. The question now is what they're allowed to see.

The decision

Last week, Sweden's Riksdag voted to authorize the police to use AI-powered real-time facial recognition in public spaces. The new regulation is expected to come into force on July 1st, 2026, alongside amendments to the Public Access and Secrecy Act. The law is tightly scoped: use must be "proportionate and absolutely necessary," court authorization is required as a rule, and in urgent cases police have 24 hours to obtain retroactive approval. Permitted uses cover kidnapping, human trafficking, serious crimes with a maximum sentence of four years or more, and imminent threats to life. The Swedish Authority for Privacy Protection (IMY) will supervise the program, and a fundamental rights impact assessment must be completed before the technology is first deployed. This is not a surveillance free-for-all. But it is a threshold crossed.

The case for it

The government's argument rests on a real and documented problem. Over the past three years, 23 bystanders have been killed and 30 wounded in gangland shootings in Sweden. These are not abstract statistics — they're people waiting for buses, sitting in parks, walking past the wrong address. The governing coalition — the Moderates, Christian Democrats, Liberals, and Sweden Democrats — argued that real-time identification could become a decisive tool in breaking the speed advantage that organized criminal networks currently hold over police response. The technology isn't banned outright by EU law either. The EU AI Act permits certain law enforcement uses of real-time biometric identification under narrowly defined circumstances, subject to judicial and regulatory safeguards — and Sweden has framed this law as operating precisely within those exceptions. If the safeguards hold, the argument goes, the benefit is measurable and the harm is narrow.

The case against it

The critics are not wrong either, and their concern isn't paranoia. The Centre Party, the Left Party, and the Greens all raised objections during the parliamentary process, citing privacy risks and calling for stricter limits — with only the Centre Party formally proposing rejection in full. Their worry is one that surveillance scholars have documented across every jurisdiction that has introduced similar technology: mission creep. What begins as a tool for murder and trafficking investigations has a documented tendency, over time and under political pressure, to expand in scope. The law as written today may be narrow. The political appetite that passed it — in a government relying on the far-right Sweden Democrats ahead of a September general election — suggests that the floor of what counts as "necessary" has every incentive to drift downward.

Where I land

Sweden's gang violence problem is real and the people most harmed by it are not wealthy or powerful. Dismissing this law as authoritarian theater misreads the genuine desperation behind it. But I'm more troubled by the political context than the text of the law. Proportionality frameworks erode when the people setting the dials have electoral incentives to look tough. Sweden has a strong privacy regulator and a functioning judicial system — but it also just handed a powerful new capability to a police force that, a few years ago, was fined for illegally using Clearview AI. The technology arrived before the culture of restraint did. That gap is what worries me.

Over to you

Sweden has chosen to use one of its most progressive values — technological leadership — in service of a very old political instinct: control. The question isn't whether the law is well-written. It's whether a well-written law is enough protection when the political winds keep shifting. Is this a sensible pragmatic response to a genuine crisis — or the beginning of something Sweden will spend years trying to walk back?

READER FAQs

What crimes can Swedish police actually use facial recognition for? 
A: Serious offenses carrying a maximum sentence of at least four years — including murder, human trafficking, kidnapping, and imminent threats to life. Routine police work and minor offenses are explicitly excluded.

Does police need a court order every time they use it? 

A: As a general rule, yes. In urgent situations they can deploy without prior authorization, but must apply for retroactive court approval within 24 hours.

Why did Sweden move ahead when many EU countries haven't? 

A: A decade of escalating gang violence — including bystander casualties — created strong political pressure. The current right-wing coalition, relying on Sweden Democrat support ahead of September elections, has made crime its defining issue.

Isn't this banned by the EU AI Act? 

A: Not entirely. The EU AI Act generally prohibits real-time biometric identification in public spaces, but permits narrow law enforcement exceptions under strict conditions. Sweden's law is designed to sit within those exceptions.

Who is watching the watchers? 

A: Sweden's data protection authority, IMY, will supervise the program. Before the technology is deployed for the first time, the police must also complete a mandatory fundamental rights impact assessment under the EU AI Act.