Facial recognition should be banned in public spaces

What's this about?

People disagree about whether towns and police should ban face-scan tools in public places. These tools can match faces to names, often without people knowing.

What supporters say

• Face scans can track who joins protests, visits shops, or walks through a town.
• This tracking may scare people away from protests or stop them from speaking freely.
• Some face-scan programs make more wrong matches for women and people with darker skin.
• A wrong match can lead police to question, watch, or even arrest an innocent person.

What critics say

• Police may need face scans during rare, urgent events, such as a search for a dangerous person.
• Supporters of limited use say judges or other leaders should approve each use first.
• They also say police should treat a computer match as a clue, not proof of guilt.
• Strict rules could limit who uses the tool, save records, and punish misuse.

The bottom line

The facts suggest police should not use face scans to watch everyone in public all the time. Rare emergency use may still make sense, but only with very strict approval and safety rules.

Facial recognition in public places should not be used for routine, real-time scanning, the evidence suggests. But the strongest conclusion is not an absolute ban in every circumstance: narrowly defined emergency uses, with strict authorization and safeguards, may still have a case.

The case for

The central concern is that facial recognition can identify people in public without their knowledge or participation. Used at scale, it gives authorities a powerful way to track who attends protests, visits public places or moves through a city. That can threaten privacy and discourage people from exercising their rights to protest and speak freely, even if they are never arrested or charged. 1

The risks are especially acute at peaceful demonstrations. Knowing that a crowd may be scanned and identified can deter people from joining it. Expert reviews have warned that the technology’s capabilities have moved faster than the laws and institutions meant to protect privacy, equality and civil liberties.

Accuracy is another major problem. Testing by the US National Institute of Standards and Technology found that error rates differ widely between systems and uses, but that many algorithms produced more false positives for some groups, including women and people of color. In a public setting, where police may scan large numbers of people, those errors can mean that innocent people are subjected to questioning or surveillance at unequal rates. 2

The consequences are not merely theoretical. Robert Williams was wrongly arrested after an incorrect facial-recognition match linked him to a theft. His case does not show how often such mistakes occur, but it illustrates the serious harm that can follow when an algorithmic alert is treated as more than an initial investigative lead. 3

There is also little reason to assume existing safeguards will reliably prevent misuse. Reviews by the US Government Accountability Office found gaps in agency reporting, policies, training, documentation and oversight. In Britain, the Court of Appeal ruled that South Wales Police’s use of automated facial recognition was unlawful because the rules did not sufficiently limit where it could be used or who could be placed on watchlists, and did not adequately address privacy and equality duties. 4

The case against

The case against a total ban begins with the possibility that facial recognition can help police in limited situations. A Metropolitan Police trial generated alerts that officers could investigate, although it also produced false alerts and required human officers to check possible matches. Some recent studies have found links between biometric surveillance and lower violent crime, suggesting that the technology may offer public-safety benefits worth examining. 5

A blanket prohibition could also fail to distinguish between constant mass scanning and a tightly controlled use during an exceptional emergency. Controlled image conditions, carefully chosen systems, small watchlists and human review might reduce some of the technology’s accuracy problems. Technical testing does not show that every system performs equally badly: results depend heavily on the algorithm, the quality of images and the particular task. 7

Some experts therefore favor risk-based rules rather than an outright ban. The National Academies has called for stronger governance, transparency, testing, accountability and rights protections, rather than declaring all uses unacceptable. A regulatory approach based on necessity and proportionality could, in theory, preserve certain investigative uses while limiting the worst forms of surveillance. 6

But the evidence for crime-control benefits remains limited. The available studies are observational: they cannot show that facial recognition alone caused any reported changes in crime, rather than broader policing or surveillance measures. Nor is there strong comparative evidence that judicial approval, narrow watchlists, independent audits and human review can consistently prevent rights violations when the technology is used at scale.

The bottom line

The evidence most strongly supports a ban on routine or indiscriminate real-time facial recognition in public spaces, particularly at protests and wherever independent oversight is weak. The risks to privacy, equality, free expression and due process are well documented, while ordinary safeguards have repeatedly proved inadequate.

At the same time, the evidence does not justify treating every possible use as equally unacceptable. The most defensible policy is a prohibition with very narrow, authorized emergency exceptions—an approach reflected in the EU AI Act, which generally bans law-enforcement real-time biometric identification in public while allowing specified exceptions under conditions and authorization (see Figure 2).

Confidence in this qualified conclusion is high. Multiple government reviews, technical tests, legal cases and expert analyses point to substantial risks. The remaining uncertainty is whether truly independent controls and tightly limited deployments could ever deliver enough public-safety benefit to outweigh those risks.