Last May, the French government launched a working group aimed at legalizing real-time facial recognition. Far from being a surprise, this announcement is part of a series of proposals put forward by state officials, along with industrial and scientific players. We publish this op-ed by Félix Tréguer, adapted from a text originally published on AOC, where he argues that facial recognition is incompatible with democratic forms of life.
In May 2025, Gérald Darmanin was getting restless. Overtaken on the right by his successor at the Ministry of Interior, Bruno Retailleau, the new Minister of Justice apparently found it difficult to hang up his apron as “France’s top cop.” So he pulled out of his hat a seemingly novel proposal: the legalization of real-time facial recognition.
The ink on the War on Drugs law, with its array of new police surveillance measures, was not yet dry— the Constitutional Council would rule on it a few days later—but the minister was already on to his next move. After his announcement, his office would confirm to AFP that a working group was about to be launched to “create a legal framework” so as to “introduce this measure into our legislation.” According to the minister, who in 2022 said he opposed facial recognition, “using technology and facial recognition are the solutions to drastically combat insecurity.” A few days later, his rival Retailleau would follow suit, calling for “highly regulated” use of real-time facial recognition in the context of criminal investigations.
A political project
The context of heightened political competition on the right could lead one to believe that this is yet another trial balloon with no future. In fact, facial recognition is a political project that has long been embraced by Emmanuel Macron’s governments, but has been repeatedly postponed for fear of provoking an outcry among the population.
When my colleagues at La Quadrature du Net and other collectives across the country launched the Technopolice campaign in 2019 to document new police surveillance technologies and unite local resistance, facial recognition was already on everyone’s lips. At the time, the Parliamentary Office for Scientific and Technological Assessment, then spearheaded by a Macronist parliamentarian, was already calling for an experimental law to authorize its use in real time. A few weeks later, Secretary of State for Digital Affairs Cédric O gave an interview to Le Monde newspaper on the subject. In this very first government statement on the topic, O considered it necessary “to experiment with facial recognition so that our manufacturers can make progress.”
The economic stakes were then expressed candidly. It is true that since the early 2010s, facial recognition and other techniques combining artificial intelligence and video surveillance—a spectrum of applications grouped under the term algorithmic video surveillance (AVS)— have been the subject huge public spending. Through public research policies led by the European Commission or the French National Research Agency, but also via tax mechanisms such as the Research Tax Credit, startups and large multinationals such as Idemia and Thales have a significant portion of their R&D financed by taxpayers. Bpifrance (the French public investment bank) and the Caisse des Dépôts et Consignations have also taken action to help French industry structure itself to gain a foothold in these promising markets: last year, the global facial recognition market grew by 16% per year and is expected to reach $12 billion in 2028; while the market for other applications of AVS was $5.6 billion in 2023 and could reach $16.3 billion in 2028.
In 2019, Cédric O therefore proposed legalizing facial recognition on an experimental basis for the 2024 Olympic Games. But on the eve of the 2022 French presidential election, before retiring from politics to become the chief lobbyist for the tech industry, he publicly acknowledged that the political conditions for using this technology were not ripe: “Priority has been given to other issues […] given the context and the sensitivity of the subject,” he explained at the time, while also denouncing the “libertarian NGOs” that, in his view, had fueled a climate of “psychosis.”
The government then fell back on VSA applications deemed less sensitive. They were legalized on an experimental and temporary basis under the 2023 Olympic Games Act and implemented in recent months: These include detecting people or vehicles driving in the wrong direction, falls to the ground, crowd movements, fires, etc. The experiment had inconclusive results, and yet the government is now seeking to extend it as part of the next law relating to the 2030 Olympic Games. But at the end of 2022, during parliamentary debates, Gérald Darmanin, then Minister of the Interior, and his colleague in charge of sports, Amélie Oudéa-Castera, were pretty clear: facial recognition was a no-go: “The system does not in any way provide for […] the creation of a biometric identification system,” the minister of sports stated in the chamber: “The government does not want anything like that, either now or in the future.”
In the AI Act, exemptions for the police
But these reassurances were part of a double game. At the same moment, in Brussels, the French government was leading the negotiations on the AI Act. As demonstrated by an investigation by the media outlet Disclose published this winter, even though the European Commission’s political marketing around this text was based in part on the promise of a ban on “real-time biometric surveillance in real time,” France was actually putting all its weight behind pressuring other European Union member states to spare law enforcement agencies from overly restrictive regulations.
This strategy paid off. In the version of the “AI Act” that was ultimately adopted, the principle of banning real-time facial recognition was immediately undermined by a number of exemptions. For example, it is authorized to prevent “a genuine and foreseeable threat of a terrorist attack,” but also in the context of criminal investigations to find suspects for a range of offenses punishable by more than four years’ imprisonment, including sabotage. Activist activities, particularly those associated with the environmental movement, could easily be impacted.
Another concession obtained by France —a particularly chilling one—, allows police forces to use algorithmic video surveillance systems that “deduce or infer [people’s] race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation.” This means not only the detection of insignia or clothing denoting political orientation, but also the resurgence of naturalizing theories and pseudosciences purporting to reveal “race” or sexual orientation based on morphological characteristics or facial features, now incorporated into powerful automated systems designed to enact state violence.
Gérald Darmanin’s announcement of the launch of a “working group” to legalize facial recognition was therefore no surprise. It was a logical follow-up of many steps taken at the highest levels of government, in conjunction with industrial and scientific actors, to prepare the French population and minimize as much as possible the political cost of legalizing real-time facial recognition.
The Minister of Justice has claimed that facial recognition is essential to ensure the safety of the population, while trying to minimize the issues at stake: “People say that at [the main Paris airport] Roissy, it takes 1.5 hours to get through; in Dubai it takes 10 minutes; yes, but in Dubai, they have facial recognition,” he explained May, touting the added convenience that people would be entitled to expect from the widespread use of this technology. One might be tempted to remind the minister: “Yes, but in Dubai, human rights defenders are imprisoned, the regime resorts to torture and is effectively a dictatorship. Is this really a model to follow?”
Darmanin conveniently forgets that facial recognition is already a reality in some French airports and train stations. It helps him better defend his fool’s bargain: privacy and freedoms in exchange for greater convenience for those who move around in the world of flows. The minister sees this as a good deal, one that will convince “the average person” to put aside their reservations and what he sees as a widespread “paranoia about technology, civil liberties, and the issue of databases.”
Permanent, general, and invisible identity checks
In high places, however, the paradigm shift brought about by real-time facial recognition is well understood. In 2019, when a colleague from La Quadrature and I were invited to give our opinion on the “social acceptability of facial recognition” before an assembly of police officials, prefects, scientists, and industrialists at the General Directorate of the National Gendarmerie, a colonel in the Gendarmerie came to present a memo he had just published on the subject. In this document, he offered a rather lucid analysis of the place of facial recognition in the history of state identification techniques:
“The advantage of this technology is that it systematically and automatically performs the basic tasks of law enforcement, which are to identify, track, and search for individuals, making this control invisible. Subject to unbiased algorithms, it could put an end to years of controversy over racial profiling, since identity checks would be permanent and universal. Provided the algorithms are unbiased, it could put an end to years of controversy over racial profiling, as identity checks would be permanent and universal [emphasis added].”
“Invisible, permanent, and universal” identity checks? Michel Foucault was undoubtedly right when, in Discipline and Punish (1975), he alluded to the fantasy of a police force that had become “the instrument of permanent, exhaustive, omnipresent surveillance, capable of making everything visible, but on the condition that it makes itself invisible.”
If a law were to be passed to authorize real-time facial recognition, things could move very quickly: Even if, for the time being, in France, its use by the police is only legally possible after the fact, in the context of judicial investigations and only for the “Criminal Record Processing” database (the TAJ, which contains nearly 10 million photographs of faces), the technical infrastructure enabling real-time use is already in place. First, the sensors: around 90,000 video surveillance cameras placed on public streets and roads across the country, forming as many geolocated points dedicated to the collection of facial images. Next, centralized databases of ID photos linked to civil status data: in addition to the TAJ database, most immigration-related files now include photographs of faces that can be processed by algorithms. This is also the case for the ” Secure Electronic Documents ” (TES) file created in 2016 by the Ministry of the Interior, which collects facial prints from all holders of identity cards and passports. And finally, the last piece of the puzzle: facial recognition algorithms that compare images to files, provided by private service providers such as the multinational company Idemia, whose reliability rate has greatly improved in recent years.
Refusing facial recognition
With facial recognition, our faces become index terms in police files, which means that our personal data can be automatically revealed (surname, first name, place of birth, place of residence, etc.). If its use in real time were authorized, going out in public with your face uncovered would be like harboring a forgery-proof ID card that could be read by the government at any time. Anonymity would be made virtually impossible. And, in the course of this process, our faces—which reflect our emotions, attitudes, and ways of being—would be reduced to being mere faces: eyes, a nose, a mouth, ears, and other anatomical features whose measurements, shapes, or colors could be automatically classified. As showcases of our subjectivity, they would become new objects of power, through which the state could control us.
Facial recognition is also one of the means by which fascism could take hold and endure. On that day in September 2019 at the General Directorate of the National Gendarmerie, we reminded all its promoters present in the audience why we believed it is unacceptable. Thinking that we could strike a chord with some in the audience, we told them of our conviction that if our grandmothers and grandfathers had had to live in the early 1940s in a world saturated with these technologies, they would not have been able to survive for long in hiding, and therefore organize resistance networks capable of standing up to the Nazi regime.
This counterfactual hypothesis illustrates why facial recognition is simply incompatible with the defense of democratic ways of life. In this day and age, it is not to be taken lightly.
Félix Tréguer is a researcher, member of La Quadrature du Net and author of Technopolice, la surveillance policière à l’ère de l’intelligence artificielle (Divergences, 2024).