The U.S. Fears Live Facial Recognition. In Buenos Aires, It’s a Fact of Life

Live facial recognition has been in use in the capital of Argentina for nearly a year.

In July 2019, Guillermo Federico Ibarrola was heading home on the subway when he was stopped by Buenos Aires police. The authorities told Ibarrola that he was being detained for an armed robbery that had happened three years ago in a city about 400 miles away.

Ibarrola protested. He said he had never even been to the city where he was accused of committing the crime. Still, he was arrested.

On the sixth day in police custody, he was suddenly released. The police officers offered Ibarrola coffee and dinner, and a bus ticket back home.

As it turned out, a “Guillermo Ibarrola” had potentially committed a crime, but it wasn’t this Guillermo Ibarrola. Three years earlier, in the wake of the armed robbery, Buenos Aires police entered the name “Guillermo Ibarrola” into the country’s fugitive watchlist. But there are at least two Guillermo Ibarrolas in Argentina, and the system ended up registering the innocent Ibarrola as the suspect. Three years later, as he made his way home, Buenos Aires’ recently instituted real-time facial recognition system flagged him to authorities.

Though it’s becoming common around the world, live facial recognition — the practice of matching every face in live security footage against a watchlist — remains highly controversial. In January, the New York Times published a story about London’s adoption of live facial recognition systems, warning that the West is heading toward a future “of constant surveillance.”

He had never even been to the city where he was accused of committing the crime. Still, he was arrested.

But that future is already here for the people of Buenos Aires, Argentina. Last April, the government of Buenos Aires announced that the city’s subway cameras would be connected to a system called the Fugitive Facial Recognition System. Three weeks later, the system was operational. For almost a year, the residents of this city of 3 million have lived under the surveillance of live facial recognition, with some individuals put on a watchlist even for minor crimes like theft.

In cases like Ibarrola’s, Argentinians have also seen what happens when facial recognition systems fail. Last June, a man was wrongly detained under similar circumstances and spent hours at a police station trying to convince officers he didn’t commit an armed robbery 17 years ago.

Buenos Aires’s facial recognition system was pitched as a way to capture wanted individuals from the country’s national fugitive database, called Consulta Nacional de Rebeldías y Capturas (CONARC). The municipal government streamlined the legislation that enabled the adoption of the technology, passing the legislation as a resolution rather than a law.

The cameras are operated by the city’s ministry of security, and operate in the city’s train stations, both within the stations and facing the street, according to documents obtained by an Argentinian civil rights organizations called Asociación por los Derechos Civiles (ADC).

The video surveillance system is developed by a Buenos Aires-based company called Danaide S.A., which sells its surveillance technology through a product called Ultra IP. The facial recognition component of Danaide’s product is developed by Russian company NTechLab, which confirmed its partnership with the Argentinian firm to OneZero, but did not provide more information due to a nondisclosure agreement. NTechLab’s software is also being used to conduct live facial recognition in Moscow using 3,000 CCTV cameras, according to a 2018 company presentation. The company claims an accuracy rate of 80%, though that has/has not been independently verified.

The Buenos Aires live facial recognition system can function on 300 camera feeds at a time, and automatically sends a notification to police over messaging app Telegram when it finds a potential match. Since there are more than 300 cameras available to city police, the video feeds from both underground and above ground are cycled through. Danaide S.A. did not reply to a request for comment, and little else is known about how the technology has been implemented.

“I fail to see the proportionality of installing a technology with serious privacy implications for searching a list of 46,000 persons that includes non-serious crimes.”

Argentinian politicians and law enforcement often express their enthusiasm for the system. Patricia Bullrich, the former Argentinian minister of security who left office in December 2019, tweeted praises of it regularly. In July 2019, she tweeted that 9,600 fugitives had been captured at a large Buenos Aires train station, with some of those detentions made with the help of facial recognition cameras.

In recent weeks, Buenos Aires police have tweeted about seven cases in which the facial recognition system was used to capture fugitives. That includes a video posted on February 4 that claims that the facial recognition system had caught a suspect wanted by Interpol for aggravated robbery.

Data provided to the ADC by the Buenos Aires government in July 2019 suggest that between April 24 and June 4, the facial recognition system identified 595 people to police. Of those 595 people, five were false positives, meaning that the identified individual was not the correct person, nor in the database. ADC notes that there’s no way to verify the data supplied by the Buenos Aires government.

Buenos Aires officials did not respond to a request for comment.

Buenos Aires isn’t the only city in Argentina using live facial recognition technology. Tigre, a city about 20 miles north of Buenos Aires, has 2,000 cameras feeding footage to a central command center operating NEC software. That footage is also used to track license plates, crowds, people laying on the ground, and loitering, according to a 2014 NEC presentation. The city has been using NEC’s surveillance technology since 2011, according to a company case study.

Civil rights organizations like the ADC object to the widespread use of facial recognition in Argentina, especially since it’s been implemented with little oversight or thought to privacy.

ADC argues that the legislation that enabled the implementation of facial recognition technology in Buenos Aires left too much room for interpretation, and could lead to abuse. ADC filed a lawsuit in October 2019 against the government of Buenos Aires, declaring the facial recognition system unconstitutional.

“When we asked for privacy impact assessments or other human rights impact assessments, they’re not doing that at all,” says Leandro Ucciferri, policy analyst and researcher at ADC. “They’re not carrying out the studies around necessity, proportionality on the risks that this technology can bring to human, social, economic, and political rights.”

The ADC isn’t the only organization concerned with how the facial recognition system and CONARC have been implemented in Argentina. In May 2019, a United Nations report called the legitimacy of the CONARC database into question, saying that the database contains crimes as low-level as simple theft. Nearly 30% of entries don’t even list what crime the fugitive is wanted for.

“I am aware of the need to arrest persons who are suspected of having committed crimes and bring them to justice, but I fail to see the proportionality of installing a technology with serious privacy implications for searching a list of 46,000 persons that includes non-serious crimes and is not carefully updated and checked for accuracy,” the UN report says.

By any technical account, Buenos Aires’ facial recognition actually worked in the case of Guillermo Ibarrola. It matched the face of a person placed on a watchlist to a stream of video it received.

But as activists like the ADC say, it’s not just the accuracy of the face matching system that can lead to wrongful incarceration, but human errors inserted into the system in ways nobody might foresee.

That means that similar clerical error could happen in London’s live facial recognition system, or anywhere else in South America, Europe, or Asia where the technology is deployed. A 2018 NTechLab presentation showed a map where the company was already selling live facial recognition. The map listed more than 100 customers in 20 countries, across five continents.

Senior Writer at OneZero covering surveillance, facial recognition, DIY tech, and artificial intelligence. Previously: Qz, PopSci, and NYTimes.