General mobilisation against the legalisation of algorithmic video surveillance!

Posted on

During the end of year holidays, and without fanfare, the government prepared the introduction of a new legislation regarding the 2024 Olympic and Paralympic games. As expected, this draft law pave the way for the generalization of algorithmic video surveillance. While using the fallacious argument of time limited experiment, such technologies are already illegally used in many cities in the country with total impunity. Using the forthcoming sport competitions as justification, the government and many representatives call themselves advocates of the French industry and use this position to allow thos companies to durably expand on the video surveillance market.

What is automated or algorithmic video surveillance ? It can be defined as the automation of the analysis of CCTV images thanks to a software that produce notifications when it detects an event that it has been trained to recognize.This analysis work was previously done by humans (municipal agents within urban supervision center ou security agents within supermarket pr private establishments). These softwares are based on so called “computer vision” algorithms, a technology built on statistical learning that makes it possible to isolate meaningful information from static or moving images. In order to isolate these informations, algorithms are trained to automatically detect, through video streams from CCTV cameras, certain categories of objects (trash, bag), people (lying on the ground, graffiti artist, static person) or events (crossing a line) for instance.

Such automated CCTV cameras lead to massive police surveillance of public space. It analyses behavior to detect the ones supposed to be considered as not “normal”. It implements algorithms that can track individuals on the streets if the police consider them “suspects”. In a near future it is will mean real time identification through facial recognition and massive usage of video based fines. It also means millions of euros from public funding for dangerous technologies deployed without real debate.

To make sure this future society never becomes real, we have to strongly reject these rules. Be ready for mobilization !

For more than three years, we have documented the illegal experimentation of automated video surveillance in French cities and villages. We are even suing in courts big cities like Marseille or small villages like Moirans (Isère) in order to denounce how illegal these artificial intelligence surveillance technologies are. All that time, the government and industrial providers have been knowing to be out laws so they tried to modify the law in order to make their activities legal. In 2020, the « global security law » was close to be used as a lawful ground. But due to fear of the opposition members, the government avoided making a final decision. Unfortunately, here we are again…

The Olympic games opportunism

It is not a coincidence if the government uses the 2024 Olympics as an occasion for proposing a legal framework on these technologies. The Olympics are considered as a «mega-event», which exceptional dimension allows the acceleration and implementation of extraordinary regulations . As seen in the previous editions, such events are a pretext for legal innovations in security matters. Laws related to the Olympics always create a framework for strict policing, together with the militarizing of the public place as well as an increase in the use of surveillance.

Researcher Jules Boykoff compares this strategy to Naomi Klein’s “shock doctrine”: Governments use a trauma, a natural disaster or a tragedy to privatize and deregulate. As for the Olympics, they emerge as an accelerator of exceptional policies but this time relying on a festive event during which laws can be temporarily suspended. Such phenomenon allows the adoption of new legislation that would be hard or even impossible to pass in normal time.

For example, the Brazilian government used the 2016 Olympics in Rio to perform quasi-military operations in favelas and to evict persons from their housing. Similarly, the Japanese government used the Tokyo Olympics to pass an “anti-conspiracy” law, which was actually planned for long in order to bring into line activists and unions. Previous governments had failed to pass such a legislation three times. This law has been highly criticized, including by the United Nations, as it undermined civil liberties as well as it gave big surveillance power to the state.

Finally, sport mega-events are often qualified as “security performances” or “security minded showcase” as they give place to trial situations and monitoring laboratories. Also, governments rely on these exceptional events to sugar coat the severity of this surveillance measures and make them more acceptable by the opinion.

Paris in the race

The choice of the French government to introduce automated video surveillance trials in the law for the 2024 Olympics respond perfectly to this scheme and logic of manipulation of mega-events. Although no public and scientific study exists on this subject and no need has been precisely identified, directives regarding automated CCTV surveillance are depicted by the government and some representatives as mandatory for the security of this event.

In the impact assessment, government lazily takes as example the detection of abandoned parcels : “Rather than waiting a report by patrol officers or localisation by operators in charge of viewing thousands pictures of video surveillance, this legislation would save a precious time” and “only the use of an algorithmic processing is able to to report in real time this risky situation and allow implied services to analyse it and take action if needed”. Here is the only justification for the introduction of an algorithmic surveillance of public space that government intends to make durable, and which potentially involves millions of people.

In reality, experimentations are part of a larger political agenda and comply with wishes expressed for years by industrial sector and police organisations to massively use these devices. Therefore it’s certain that these tools won’t be dropped after experimentation period, in the same way that “black boxes ” of intelligence services or derogatory rules of “emergency state” of 2015 were made durable, although they were supposed to be temporary and exceptional. Some representatives prominent on this subject, as Philippe Latombe, are already talking about another law, more generalistic, to come on the same subject for year 2023.

A freepass for the automated CCTV surveillance market

In order to allow automated CCTV surveillance development, government planned an article 7 inside the draft law, which provides for an experimentation process divided in several steps until June 2025. Behind this apparent procedural structure, article 7 constitutes in reality a booster for selling automated CCTV surveillance softwares to state and local authorities by private sector.

If the law is portrayed as concerning the Olympics, in reality, algorithms will be deployed n a huge number and various kind of events during the time frame set for experimentations: music festivals, marathon races, outdoor shows, Christmas fairs or 2023 rugby World Cup. Then, many “sports”, “entertainment” or “cultural” event will be the playgrounds of these algorithms. By the way, these events scope is also very broad as the captured pictures of the surroundings of the events places, as to means of transportation which connect them (inside vehicles and circulation ways) will also be taken up for the experimentation.

First of all, experimentations are supposed to meet a double requirement that perfectly respond to the expectations of the CCTV surveillance market. First, the purpose must be “to ensure security of sport, entertainment or cultural events which, by their magnitude or their circumstances are particularly exposed to risks of terrorist attacks or great security risk for the population “. Secondly, the algorithms shall “detect, in real time, predetermined events likely to show or reveal these risks and to report them in order to implement necessary actions ” for various types of security agents (national police forces, local police, fire and medical departments and internal security departments of transportation authorities). Processed pictures are those coming from surveillance cameras or drones, the drones being allowed in France since last year.

As with many surveillance systems, the government pretends such technology is needed to help to prevent risks. And as usual, such risks are so broadly defined that they can include a large range of situations. Also, the ‘event detection’ feature fits perfectly with the core product of devices engineered and sold by automated CCTV surveillance: guess and classify the “risky” behaviors in public space and report those who would require police attention and intervention. So, in the law, “algorithmic processes ” would allow to detect in the way we live, move or walk any behavior that should or not trigger an alert.

In the law impact assessment, the only “social impacts ” raised by the government would actually be advantages: automated CCTV would “reinforce” security and would guarantee “preservation of anonymity in public space and less privacy invasion” thanks to the lack of facial recognition – no kidding ! – and finally the law requirements would ensure “tool neutrality“. In its opinion, the “Conseil d’État” (the highest administrative jurisdiction body with power to block or change laws according to civil rights abuses) follows the same argumentation as the one promoted by the industry, stating that algorithms function associated with CCTV is limited to a “removal of doubt”. Then, the Conseil d’État minimises this software influence in the decision process made by security agents, who could disclaim their responsibilities under the umbrella of a machine. The CNIL (the French data protection authority) has not been seen or heard on the matter, merely thanking itself that “garanties” required in a previous guidelines are included in the draft law. Nut let’s remember that such guidelines called for legalization of automated CCTV..

We have denounced such dangers many times: these tools are all but neutral. In the industry common practices, such events can consist in “suspect behavior detection “, “area wandering” (being static in public space) , “virtual line crossing ” (which means someone entered a defined zone), tracking people, abandoned items, riots or theft detection. Behaviors deemed “suspects” are not a tangible reality, but are only incarnation of subjectives and discriminatory political choices, focused on people spending the most time in the streets: for exemple, beneath the “area wandering ” detection feature is hidden the hunting of beggars. Either it is human or algorithmic driven, image interpretation is always ruled by social and moral vision, and adding a software layer will not change anything. In the end, CCTV surveillance automates and amplify police decision process, to increase coercion and repression powers.

In the same strategy of minimizing surveillance, the government also claims that these devices would not process biometric data (which is false as we already explained, and as the Défenseur des Droits – a civic rights institution – recently recalled) and that it will not run facial recognition. By showing itself as renouncing for the moment to these specific technologies -that bring to the population a singular perception – the government plays a political and communication trick to give automated CCTV surveillance a less dangerous appearance. The message consist in stating that algorithmic CCTV is relying “only” on a purely technical image analysis which would only “assist” human operators. Once again, the government and the Conseil d’État are directly inspired by the basic promotion speech of the automated CCTV industry. This surveillance cannot be considered as less dangerous than facial recognition as it affects the same body intimacy. In other words, facial recognition is only one of the many use of automated CCTV and is very likely to be legalized to identify people in real time in public space, in one of the many security-related law to come. As of 2018, representatives like Christian Estrosi, the mayor of Nice, were promoting such facial recognition usage to legitimate the implementation of automated CCTV surveillance in their towns.

The State at the helm

The draft law would allow conception and deployment of surveillance algorithms through a administrative process divided in several steps, all of them being managed and supervised by the government. Also, this process contains the less limitations possible.

First of all, as with non algorithmic CCTV, government agencies or local authorities will have to require for an autorisation to the local “Prefet’ (the administrative regional authority). These autorisation requests will have to identify in which precise event the experimentation will take place (for exemple the Paris marathon in April 2023, or the big music festival “Les Vieilles Charrues ” in July) and which algorithm will be tested (which “event” to detect, on what purpose the use of automated CCTV surveillance is justified, who can access it, how is it funded). A risk assessment must be attached to this request. However if this kind of formal document is usually required for privacy or personal data cases, it has proven for long its lack of effectiveness to protect civil rights: these assessments are usually poorly developed and almost never succeed in practice in preventing the deployment of systems that could potentially be deemed illegal by justice.

To build such softwares, the government and local authorities will have three options: concieve it by themselves, have it done by a third party, of buy it out off the shelf. The last two options are likely to be favored, as it would be cheaper and quicker to commission contractors working on the technology for years rather than doing internally. And for the automated CCTV surveillance industry – including big names like Thales or Idemia or small start-ups like Briefcam, Aquilae, XXII, Two-I or Datakalab- it will be a golden opportunity! This law will allow them to increase deployment of their products by providing a legal framework that was lacking until now and by rerouting all of their “computer assisted vision” technological developments toward security goals.

As for the crowd that will be analyzed through these surveillance system during the many events in which tests will be implemented, well they will serve as Guinea pigs: the draft law provides that pictures coming from CCTV in those areas can be used as training data if their retention period is not expired. However, in the impact assessment study it is said that “if reuse of these data is needed to adjust processing parameters, they can be stored longer that original retention periods, in the limit of the time of experimentation and for this use only, in order to improve algorithms training“. Concretely, it means that automated CCTV surveillance companies will benefit from a huge volume of images to train their algorithms. Filmed citizens would become their lab rats.

On the technical aspect, the draft bill provides that the software development meets expectations that are fully out of step with the reality of how such softwares actually work. We will come back in details in a forthcoming article.

If the requirements are met, a legal authority (not yet defined) will give a compliance certificate but the law is silent on the methods and criterias that will be used to deliver such certificiate. In the impact analysis, it is said that the government will probably ask help to a “specialised certification cabinet”. Maybe another private actor? Anyway, the only prerequisite for the companies is to “give proficiency and sustainability insurances, and to provide complete technical documentation “.

Once the software is set up, the “préfet” (the regional state autority) will give formal authorization on the use of algorithms to the security departments who would request it. The “préfets” will also be the only ones able to decide when to stop the usage of software or to renew the authorization, on a monthly basis, if requirements are still met. What strikes us is the remarkable absence of the French data protection authority ( “CNIL”). Evicted from the process, CNIL’s powers are extremely limited and its opinion is never binding, from the experimentation authorization (only a consultative opinion), to the prefet’s decision to implement the system live (the decision is only notified to the CNIL ) or even in the experimentation monitoring process (authorities inform CNIL at their own will).

This lack of independent control is not a surprise and can be explained with the continuous loss of power of French Data protection authority (DPA) Since the beginning of 2000’s, the french DPA has been put aside from state surveillance matters to focus more on economic problematics. All these past years, the CNIL has been silent and weak towards the increasing market of algorithmic CCTV and has preferred to endorse a indulgent attitude by talking openly with those companies. This confirms that the CNIL has now become an economic regulatory authority for the market of surveillance.

The struggle begins

The government had this project lined up for a long time. We do not demand to improve or patch this law up: we simply want the rejection of this article 7 for the surveillance society project it embodies.

The automated CCTV surveillance changes the dimension of mass surveillance. By allowing the State to analyse, classify and evaluate all movements and behaviors of each individual in the public space and by giving the police ten-fold increased decision powers for its decision making, this technology transforms our relation to public space and multiplies the control and repression capacities of sucurity agents.

We will struggle for this legalization of automated CCTV surveillance never to see the light of day. We will need you to show how civil society is opposed to these technologies, and to call out to the members of parliament. The battle starts on 18th January at the Senate, before the draft bill is transmitted to the National Assembly. We will shortly get back at you with new analyses and tools to take action with us!