Entry Point#3: necrotechnics

Reaction to “Sortir des nécrotechniques des frontières” (Getting out of border necrotechnics), Hubert Guillaud, Dans les algorithmes, 29 September 2026

In his letter no. 86, Hubert Guillaud devotes a long article to the work of the lawyer and anthropologist Petra Molnar. He relays an op-ed published in Tech Policy Press, in which Molnar proposes a word to name what debates on border surveillance leave aside: the dead. By border necrotechnics, she means:

the automated systems, databases and surveillance architectures through which States and companies decide which lives will be exposed to death, deportation or endless surveillance, at the border and beyond. For her, the deaths caused by these devices are neither a malfunction nor a side effect: they are “the normal functioning of the system”.

I do not wish to discuss the facts reported in the article. There is, however, a diagnosis that is made and that seems to me excessive on two points: the intention attributed to the providers of these tools, and the powerlessness of the law and, more generally, of regulation.

The providers’ intention

Saying that death is the normal functioning of the system amounts to attributing a purpose to it. Yet intention lies first (and/or also) with whoever decides to deploy, and therefore to use. Providers of artificial intelligence systems are generally not the users of their own systems. I will not take up the image of the hammer, which serves to build as well as to kill, so worn out is it, but the reasoning is close. There is also the whole question of the non-neutrality of a technique, which I have already addressed and which should shed light on this reflection.

Molnar’s remarks are mainly aimed at companies. Her definition mentions States in passing, but it mostly describes a responsibility of private actors. However, by placing providers and users in a single system endowed with its own purpose, one puts on the same footing the one who designs a tool and the one who chooses to point it at people.

Some AI tools are configured in questionable ways, sometimes without any justification, but criticism should avoid focusing on providers, because it misses what Molnar’s text highlights in spite of itself: the increasingly concrete will of certain users to mobilise AI, and technology in general, for deadly ends. These tools are offered and used because some actors do not bother with certain “ethical” questions. The use of Palantir by the US authorities in hunting down migrants seems to me an illuminating illustration of this situation. The end justifies the means and, what is more, one no longer bothers to reflect on that end….

Nevertheless, placing responsibility on the provider alone deprives the reflection of depth and, above all, of legal grounding. It is on this terrain that this account raises the most difficulties.

The AI Act

Molnar relies on Regulation (EU) 2024/1689 on artificial intelligence to conclude that it seems impossible to reform the deployment of these tools. According to her, the text

on paper classifies border control systems as high-risk systems, while excluding from its scope applications related to law enforcement, national security and migration control. It would entrust the oversight of their effects to self-assessment and internal compliance, with no real prospect of redress.

This reading seems to me inaccurate, and on certain points contrary to the text.

First, the scope. Article 2(3) excludes AI systems used exclusively for military, defence or national security purposes, and recalls that the regulation does not affect the competences of Member States concerning national security. Law enforcement and migration control are not excluded: they appear in Annex III, points 6 and 7, among the high-risk areas. What escapes the regulation therefore falls within matters governed by national law.

This is not a total exclusion from regulation. It is the States that will adopt, or not, the laws authorising this or that use. The point is far from anecdotal and it is the central point of my criticism.

Next, control mechanisms are not absent. There is a whole series of mechanisms provided for by the Regulation. For most of the high-risk systems in Annex III, there is a conformity assessment to carry out, based on the provider’s internal control. The criticism of self-assessment is therefore not inaccurate, but it is not enough, in my view, to conclude that no reform is possible. Moreover, on migration, asylum and border control management, the high-risk AI systems concerned number 4 but are “authorised” insofar as their use is also permitted under applicable Union or national law.

So it is not quite a desert. On these subjects there is legislative control that should not be neglected or set aside.

The legislature keeps control

In these areas, it is therefore the political, and thus legislative, institutions that decide on the use of this or that device. It is the parliamentary assembly that settles the question of whether we accept, as a society, interferences of this nature. Article 5(5) of the regulation illustrates this well: a Member State may provide in its national law for the possibility of authorising real-time remote biometric identification in publicly accessible spaces. If it does not, the AI Act’s prohibition remains in place.

We have therefore retained democratic control over these choices. The impossibility of reform that Molnar describes must therefore be strongly qualified, since we have the possibility of opposing this type of use through national laws.

I had defended a similar idea in a note on AI systems in human resources. As the AI Act’s entry into application approached, recruiters were worried about whether they could continue to use systems testing candidates’ resistance to stress. In my estimation, these systems should not be allowed, because they amount to legitimising a certain form of “torture”. On another subject, our institutions voted rules that rule out emotion recognition in the workplace, except for medical or safety reasons.

Watching the legislator too

The role of politics keeps growing more complex. The recent history of the regulation of these devices, or the period of the health crisis, shows that political institutions tend to resort to disproportionate means to achieve certain objectives.

The Court of Justice of the European Union and the European Court of Human Rights share the criteria governing any interference with citizens’ fundamental rights: it must be necessary in a democratic society and proportionate to the aim pursued. These criteria are regularly used to strike down national provisions that went too far. Belgium experienced this with the annulment of data retention laws, for example. We must therefore remain attentive to the gesticulations of Big Tech, but also to the attitude of our national regulators and legislators.

In this respect, Molnar’s op-ed leaves aside another part of the regulation: the practices whose risk is deemed unacceptable and which are, as such, prohibited (Article 5). On analysis, most of these practices did not need an AI regulation to be prohibited. The GDPR, Directive (EU) 2016/680 on data processing by law enforcement authorities, the Charter and the European Convention on Human Rights already formed a sufficient arsenal to regulate such behaviour.

A closer reading shows, however, that the regulation above all organised the conditions under which public authorities may derogate from these prohibitions. Real-time remote biometric identification in publicly accessible spaces, prohibited in principle, remains permitted to search for victims of abduction or trafficking, to prevent an imminent threat to life or a terrorist attack, or to locate a person suspected of certain serious offences. The disproportion I mentioned above risks resurfacing here, lodged in the exceptions to prohibited practices invoked by certain public authorities through laws passed in haste.

Our society is going through turbulence. It is manifold and, since the arrival of generative systems and other complex technological devices, takes on a new configuration and democratic turn. Criticism of AI system providers is useful and necessary, but sometimes remains a little easy. Criticism of our own uses or of our own institutions requires more effort, because it confronts us with our own contradictions. Placing responsibility where the law places it, with those who authorise and with those who use, is my reflex as a lawyer attached to the rule of law.

This must not prevent us from keeping an eye on those who authorise and those who use. Those people are also you and me. It is a stance I continue to take, and it may have to be revised if some stop playing their part.

Note on the use of a generative device

Given the subject, it seems only right to me to specify that the ideas, references and outline of this text come from my personal notes. I asked Claude (Anthropic) to put them into shape. I reread and amended the result before publication.

These notes are part of Duality, my digital notebook. If you would like to receive them directly in your mailbox, you can subscribe to the newsletter.