Low-cost authoritarianism - What the Argentine case teaches about AI surveillance

Introduction: a report, and the right question to ask of it

On 18 August 2026, Amnesty International released a report on the expansion of mass surveillance technologies by the Argentine government1. The documented facts are these: between 2024 and 2025, the administration of President Javier Milei acquired at least 1.2 million dollars’ worth of surveillance tools — platforms for collecting and analysing data from social networks, websites, the dark web and public databases; facial recognition software purchased from companies known for exploiting biometric data; drones and ground stations equipped with automated tracking, thermal cameras and real-time image transmission2. The report, based on twenty-one interviews and on the analysis of public procurement documents, describes the constitution of databases of protesters and documents one precise effect: journalists, activists, pensioners and migrants state that they self-censor on social networks and reduce their participation in gatherings for fear of being identified3.

The temptation, faced with such a document, is to read it as the chronicle of a distant aberration — one more episode in the record of a government whose polemical style occupies international attention. That reading would be doubly mistaken. Mistaken as to scale, first: 1.2 million dollars is not the budget of a totalitarian apparatus, it is that of a piece of municipal equipment. Mistaken as to distance, second: nothing in the instruments described is unavailable on the European market, and nothing in the procedures employed is inconceivable in a member state of the Union.

The analytically fruitful question is a different one: through which legal channel does a constitutional democracy equip itself with a surveillance infrastructure? The Argentine case answers it with unusual clarity, because the acts are public, dated and consultable. The thesis defended here is that this channel is not that of the exception — the state of emergency, the special statute, the identifiable tipping point — but that of ordinary administration: ministerial resolutions, adopted without parliamentary passage, justified by public security purposes none of which is in itself illegitimate. We shall examine first the Argentine arrangement as the texts constitute it (I), then the three displacements it performs (II), then what “low cost” changes in the classical analysis of state surveillance (III), before turning the mirror towards Europe (IV).

I. The arrangement: what the texts say

A. Three administrative acts

The infrastructure documented by Amnesty International rests, in essence, on three acts of infra-legislative rank, all adopted by the Ministry of Security.

Resolution 943/2023, adopted in December 2023 in the government’s first weeks, establishes the protocol for maintaining public order in the event of road blockades — the “anti-picket protocol”, named after the roadblocks that structure Argentine social protest4.

Resolution 428/2024, of 27 May 2024, frames “cyberpatrolling”: it authorises the federal police and security forces to conduct preventive tasks in digital space using open sources and publicly accessible information, for a range of offences running from threats to terrorism by way of fraud and trafficking5.

Resolution 710/2024, signed on 26 July 2024 and published in the Official Bulletin on the 29th, creates the Artificial Intelligence Unit Applied to Security (UIAAS), attached to the cybercrime directorate. Its mission deserves to be cited in the terms of the text: the prevention, detection, investigation and prosecution of crime through the use of artificial intelligence. Its functions include patrolling open social networks, applications, websites and the dark web; processing large volumes of data to establish profiles of suspects and links between cases; and analysing online activity in order to detect threats, identify the movements of groups and predict disturbances6.

A fourth act completes the edifice in 2025: Decree 383/2025, which repeals the organic law of the Federal Police in force since 1958 and inscribes cyberpatrolling among the institution’s explicit preventive powers — without judicial authorisation where it bears on open sources7.

B. What is remarkable about this list

What is striking in this enumeration is not what it contains but what it does not: a statute. None of the founding acts of the arrangement passed through the Argentine Congress. The creation of a unit tasked with profiling and predicting disturbances, the authorisation given to security forces to patrol digital space, the recasting of the Federal Police’s own statute — all of this proceeds from ministerial resolutions and decrees, that is, from instruments whose adoption requires neither public debate, nor a vote, nor an adversarial statement of reasons.

What this finding does not establish must be stated at once. It does not establish the illegality of the acts, which rest on the ministry’s existing competences. Nor does it establish repressive intent: each text speaks the vocabulary of crime-fighting, and the stated purposes — fraud, trafficking, terrorism — are those of any modern police force. The finding is procedural, and that is precisely why it generalises: an infrastructure each brick of which is administratively banal can be erected without any single moment calling, by itself, for democratic contradiction.

II. Three displacements

A. From statute to regulation

The first displacement has just been described: the framing of surveillance leaves the domain of statute for that of the administrative act. Its consequence is not only the absence of prior debate; it is also the asymmetric reversibility of the arrangement. A statute is repealed by a statute, at the end of a public procedure; a resolution is amended by a resolution, in a day, in either direction. The infrastructure so constituted is at once fragile in law and durable in fact — for if the text can disappear in a day, the files constituted, the software acquired and the operational habits installed do not disappear with it.

B. From investigation to patrol

The second displacement lies in one word of the May 2024 text: preventive. The classical judicial investigation starts from an offence and works back towards its author; it is bounded by a case file, controlled by a judge, limited in time. The digital patrol proceeds in the opposite direction: it starts from a population — the users of social networks — and searches its activity for indicators of risk. The ground invoked for dispensing with judicial authorisation is that the sources are “open”: what is publicly accessible may, it is said, be publicly collected.

That reasoning meets an objection which doctrine formulated long ago and which the Argentine case illustrates: the accessibility of an isolated datum says nothing about the intrusiveness of its systematic collection. A passer-by may read a placard at a demonstration; a system that photographs every placard, identifies every bearer and stores the whole in a searchable database performs a processing operation of an altogether different nature, though each datum taken in isolation was “open”. Aggregation changes the object. This is precisely what the report documents when it describes databases of protesters8.

C. From event to infrastructure

The third displacement is the one summarised by Amnesty International researcher Matt Mahmoudi: these technologies now form an integral part of the infrastructure of state control9. The term infrastructure should be taken seriously, for it designates a particular mode of existence of constraint. An exceptional measure is visible: it has an entry-into-force date, a perimeter, adversaries. An infrastructure is not visible: it is there, available, and its use requires no remarkable decision. Surveillance ceases to be an event and becomes a permanent capability whose activation is a matter of settings.

It is at this point that the effect documented by the report acquires its theoretical significance. Self-censorship — people who give up posting, demonstrating, signing — does not presuppose that surveillance is actually exercised on each of them; it presupposes only that each knows it could be. The product of a surveillance infrastructure is not the information collected; it is the modification of the behaviour of those who know themselves collectable. An earlier contribution published on this site argued that the protection of the inner forum constitutes the infrastructure of cognitive sovereignty; the Argentine case supplies its exact negative — an infrastructure whose effect is to inhibit the formation and public expression of judgment10.

III. What low cost changes in the analysis

The political analysis of state surveillance was built on capital-intensive cases: the Chinese systems, the programmes of the American agencies, the Israeli intrusion platforms. These cases installed the idea that mass surveillance is the business of great powers or wealthy clients.

The Argentine figure invites a revision of that premise. A budget of 1.2 million dollars — which the report gives as a minimum — is today sufficient to equip a ministry with social network monitoring, facial recognition and automated aerial surveillance11. The cause is well known: these technologies have become market products, sold off the shelf by a global commercial ecosystem. The survey conducted by Steven Feldstein as early as 2019 established that AI surveillance technologies were already deployed in at least seventy-five states, and that 51% of advanced democracies used them — the index, by construction, not distinguishing legitimate uses from others12.

An economic-political consequence follows which should be stated plainly: the barrier to entry of mass surveillance is no longer budgetary or technical; it is exclusively legal and institutional. What separates a state that surveils massively from one that does not is no longer its wealth or its engineering competence — it is the density of the procedures that stand between a minister and a purchase order. Where those procedures reduce to the signature of a resolution, capability follows will within weeks. The fifteenth consecutive annual decline in internet freedom measured by Freedom House suggests that this slope is not a hypothesis but an observed trajectory13.

IV. The European mirror, without complacency

A. What the Union prohibits

Comparison with the European framework is unavoidable, and it begins with a finding flattering to the Union. A substantial part of what Argentina has institutionalised by resolution corresponds to what Regulation (EU) 2024/1689 prohibits or strictly frames: the practices prohibited by its Article 5 — among them social scoring, the assessment of the risk that an individual will commit an offence based on profiling alone, the constitution of facial recognition databases through untargeted scraping of images, and real-time remote biometric identification in public spaces for law enforcement purposes, save for enumerated exceptions — have applied since 2 February 202514. The gap is therefore not rhetorical: it is normative, dated, and sanctionable.

B. Three reasons not to rest content

Complacency, however, runs up against three facts.

The first is French. Law no. 2023-380 of 19 May 2023 authorised, on an experimental basis, the algorithmic processing of video-protection images to detect eight predetermined categories of events, an experiment conducted on the occasion of the Paris Olympic Games; initially limited to 31 March 2025, it was extended to 1 March 2027 by an amendment to the law on transport security — an amendment whose procedural regularity was itself challenged as a legislative rider15. Here, in a European democracy, is the grammar of the Argentine case: a capability installed for an event, prolonged beyond it, through the most discreet procedural channel available. The French experiment is incomparably more framed than the Argentine arrangement — enumerated purposes, exclusion of facial recognition, provision for evaluation — but the dynamic is the same: the infrastructure outlives the event that justified it.

The second is commercial. The technologies acquired by Buenos Aires come from a global market of which European and allied companies are active suppliers. The Union regulates uses on its territory; it only imperfectly regulates what its ecosystem sells abroad. A demanding internal framework is compatible with an external contribution to the problem.

The third is legal, and it is the most uncomfortable. Argentina has benefited since 2003 from a European Commission adequacy decision on data protection — a decision reviewed with ten others in January 2024 and maintained16. A state can therefore simultaneously offer, within the meaning of European data law, a level of protection “essentially equivalent” to that of the Union, and erect by administrative act a surveillance infrastructure that European AI law would prohibit on its own territory. This coexistence does not reveal hypocrisy; it reveals that adequacy instruments measure a general legislative framework, not administrative security practice — and that between the two, the gap may lie exactly where the essential is decided.

Conclusion

The report of 18 August does not describe an authoritarian regime: Argentina remains a constitutional democracy, with a press that investigates, courts that can be seized and elections whose outcome is uncertain. That is precisely what gives the case its value. It shows that a surveillance infrastructure requires neither the end of democracy, nor an exceptional budget, nor even a statute: three resolutions, one decree and 1.2 million dollars suffice, each of these acts being defensible taken in isolation.

The political analysis of technological authoritarianism should draw from this a consequence of method. The relevant point of vigilance is not the spectacular moment — the emergency statute, the state of exception — which public opinion knows how to recognise; it is the slow, procedural channel through which a permanent capability installs itself without any decision ever taking the form of a choice of society. The question a European citizen may legitimately put to their own legal order is therefore not: are we Argentina? — the answer is no, and it is documented. It is: how many signatures would it take, here, to do what has just been done there? The quality of a democracy faced with these technologies is measured exactly there — in the number and the publicity of the acts that separate intention from infrastructure.

Bibliography

Report and documentation

• AMNESTY INTERNATIONAL, State of Surveillance: The Expansion of the Use of Mass Surveillance Technologies by the Government of Argentina, released 18 August 2026; French-language presentation by Amnesty International Belgium, 19 August 2026.
• AMNESTY INTERNATIONAL, Impact of Digital and AI-Assisted Surveillance on Assembly and Association, IOR 40/0484/2025.

Argentine normative acts

• Ministry of Security (Argentina), Resolution 943/2023, Official Bulletin of 15 December 2023 (protocol for maintaining public order in the event of road blockades).
• Ministry of Security (Argentina), Resolution 428/2024 of 27 May 2024 (guidelines for cyberpatrolling on open sources).
• Ministry of Security (Argentina), Resolution 710/2024 of 26 July 2024, Official Bulletin of 29 July 2024 (creation of the Artificial Intelligence Unit Applied to Security).
• Decree 383/2025 (new statute of the Argentine Federal Police, repealing the organic law of 1958), as analysed by Chequeado.

European and French framework

• Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence, Article 5.
• Law no. 2023-380 of 19 May 2023 on the 2024 Olympic and Paralympic Games, Article 10; Decree no. 2023-828 of 28 August 2023; Decree no. 2023-939 of 11 October 2023.
• Sénat (France), information report no. 374 (2024-2025), Vidéoprotection algorithmique, sécurité privée : après les JOP, jouons les prolongations.
• European Commission, report of 15 January 2024 on the review of the eleven adequacy decisions adopted before Regulation (EU) 2016/679.

Literature

• FELDSTEIN S., The Global Expansion of AI Surveillance, Washington, Carnegie Endowment for International Peace, Working Paper, September 2019.
• FELDSTEIN S., The Rise of Digital Repression. How Technology Is Reshaping Power, Politics, and Resistance, New York, Oxford University Press, 2021.
• FREEDOM HOUSE, Freedom on the Net 2018: The Rise of Digital Authoritarianism; Freedom on the Net 2025.
• ZUBOFF S., The Age of Surveillance Capitalism, New York, PublicAffairs, 2019.

Notes

1. AMNESTY INTERNATIONAL, State of Surveillance, op. cit. The report was presented on Tuesday 18 August 2026; its French-language presentation by the Belgian section is dated 19 August.
2. Ibid. The figure of 1.2 million dollars is given by the report as a minimum, established from public procurement documents.
3. Ibid. The stated methodology comprises twenty-one interviews and freedom-of-information requests.
4. Resolution 943/2023, Official Bulletin of the Argentine Republic, 15 December 2023. The protocol’s operational content is not detailed here; only its existence and date are relied upon.
5. Resolution 428/2024 of 27 May 2024, text published on the Argentine normative portal.
6. Resolution 710/2024, original text published on the Argentine normative portal; Official Bulletin of 29 July 2024. The functions cited — patrolling open networks, profiling, predicting disturbances — appear in the text of the resolution.
7. Decree 383/2025, as described by the fact-checking outlet Chequeado, which notes the absence of any judicial authorisation requirement for open sources.
8. AMNESTY INTERNATIONAL, State of Surveillance, op. cit., on the constitution of databases of protesters.
9. Statement by Matt Mahmoudi, Amnesty International researcher on artificial intelligence and human rights, accompanying the report’s publication.
10. “The sovereign mind and the machine”, published on this site, on the protection of the forum internum and Article 5 of Regulation (EU) 2024/1689.
11. See note 2.
12. FELDSTEIN S., The Global Expansion of AI Surveillance, op. cit.: survey covering 176 countries, documented use in at least 75 states, 51% of advanced democracies concerned; the author specifies that the index does not distinguish lawful from unlawful uses.
13. FREEDOM HOUSE, Freedom on the Net 2025, finding a fifteenth consecutive annual decline in internet freedom.
14. Regulation (EU) 2024/1689, Art. 5; the prohibitions of Chapter II have applied since 2 February 2025, a timetable which Regulation (EU) 2026/1744 did not alter.
15. Law no. 2023-380 of 19 May 2023, Art. 10; Decree no. 2023-828 of 28 August 2023 enumerating the eight categories of events; extension to 1 March 2027 by amendment to the law on transport security (2025), whose characterisation as a legislative rider was raised before the Constitutional Council. The experiment excludes facial recognition.
16. Adequacy decision concerning Argentina (2003); European Commission, report of 15 January 2024 concluding that the eleven pre-GDPR adequacy decisions should be maintained.