Members of the European Parliament endorsed on Wednesday a blanket ban on AI-powered facial recognition in public spaces.
With 335 votes in favour and 235 against, lawmakers pushed to move forward with the sweeping prohibition, rejecting an amendment that could have paved the way for law enforcement to resort to real-time biometric identification in exceptional cases.
Biometrics refers to systems that analyse biological features, such as facial traits, eye structures and fingerprints, to determine a person's identity, usually without the person's consent. Its possible use by government agencies has been often linked to mass surveillance and authoritarian regimes.
The total ban is part of a draft piece of legislation, known as the Artificial Intelligence Act, that aims to ensure the development of human-centric, ethically responsible and environmentally sustainable AI systems across Europe.
The act as a whole was backed by 499 votes in favour, 28 against and 93 abstentions during a plenary session in Strasbourg.
The regulation, which is considered a world-first attempt to rein in AI's excesses, still needs to be negotiated between the European Parliament and member states in what is known as a trilogue.
The talks, due to kick off on Wednesday evening, are expected to be intense and heavily influenced by the sudden emergence of chatbots, a rapidly evolving technology that policymakers are still trying to understand.
Brussels hopes to wrap up the AI Act before the end of the year.
The amendment rejected on Wednesday had been submitted by the centre-right European People's Party (EPP) and would have allowed law enforcement to use live remote biometrics in three different cases: the search for missing people, the prevention of a terrorist attack and the detection of criminals who are wanted under a European arrest warrant.
MEPs also voted down an EPP-drafted recital that argued the risks posed by real-time biometrics in public spaces could, in extraordinary circumstances, be "outweighed by the substantial benefits to society, persons and particularly children's safety and life."
The recital received 233 in favour and 327 against, bringing it down.
Brando Benifei, a socialist MEP who acts as co-rapporteur in the AI Act, criticised the EPP for attempting to "politicise" the issue and tabling amendments that ran contrary to the compromise previously reached at committee level.
"This tentative completely failed and I'm sure (this) will bring the EPP with a more responsible attitude back to the table," Benifiei said.
"It’s better to respect pacts. Otherwise, you lose face."
Constant and clear boundaries
The AI Act, presented in April 2021 by the European Commission, has a pyramid-like structure that splits AI-powered systems into four categories according to the potential risk they pose to society: minimal, limited, high and unacceptable.
In the original proposal, the Commission classified the use of real-time biometrics in public spaces as having an unacceptable risk to citizens and therefore strictly prohibited. The executive, however, did include three targeted exemptions for law enforcement, which the EPP appeared to echo in its amendment.
But when the file reached the European Parliament, MEPs decided to broaden the list of banned AI systems and did away with the dispensations foreseen for remote biometrics, calling the technology "intrusive and discriminatory."
The blacklist also includes biometric categorisation based on sensitive characteristics such as gender and race; predictive policing systems; emotion recognition systems in law enforcement, schools and work offices; and untargeted scraping of images obtained from the Internet in order to create facial recognition databases.
"We have seriously looked at the interests of society and our citizens in terms of privacy. And this is why we have gone one step forward than the Commission by taking away the exclusions for law enforcement," said Dragoș Tudorache, the other co-rapporteur in the file. "But we did not do it without thinking about the issue of safety and security."
Tudorache noted the compromise text would still allow the use of remote biometrics by law enforcement but only after the commission of a serious crime, rather than in real-time, and under judicial authorisation.
The possible use of biometrics by law enforcement has long been a controversial point of discussion, with many MEPs describing the practice as incompatible with democratic values. The debate has been inevitably shaped by developments in China, where the Communist Party has rolled out a massive, sophisticated network of facial recognition cameras to monitor the country's population.
"Going forward, we're going to need constant, clear boundaries and limits to artificial intelligence," Roberta Metsola, president of the European Parliament, said after Wednesday's vote.
"And here's one thing we will not compromise on: any time technology advances, it must go hand in hand with our fundamental rights and democratic values."
This article has been updated with new details about the vote.