Two days before the State of the Region Debate, the Madrid regional government approved the draft of its first Digital Administration and Artificial Intelligence Act and sent it to the Assembly to begin its parliamentary passage. The text provides for transparency, human oversight and safeguards around automated decisions. [1]
On 10 September, Isabel Díaz Ayuso took the floor to explain what kind of Madrid she governs and where she intends to take it. Artificial intelligence appeared several times, but almost always as a tool that is already working or about to do so. Rarely as a political question. Not because AI is still far away.
In 2025, 37.9% of Spain’s population aged 16 to 74 had used some form of generative AI in the three months before the survey. Among people aged 16 to 24, the figure reached 75.6%, and 59.3% used it for formal education. [2] In the first quarter of 2025, 30.1% of Madrid companies with ten or more employees were using AI technologies, compared with 21.1% across Spain. Madrid ranked first among the autonomous communities. [3]
The Administration is not waiting either. By February it had counted 220 artificial-intelligence use cases, 96 of them already deployed. Healthcare accounted for 130 projects, with 77 in operation. [4] During the Debate, Ayuso also announced 30 AI agents for Cuenta Digital, a €50 million Artificial Intelligence Factory, generative tools for public-sector staff, solutions for EducaMadrid, a regional digital twin and new healthcare systems. [5]
Madrid is doing AI. What is striking is how little Madrid discusses politically what it means to do so.
When the algorithm enters the hospital
The 2026–2028 Digital Health Strategic Plan provides for investment of more than €336 million. Announced projects include SERMAS Responde, Personaliz@ and PREVEN-IA. The latter relies on digital twins that will integrate clinical, genomic, laboratory and medical-imaging data to simulate scenarios and support therapeutic decisions under professional supervision. [6][7]
CRON-IA aims to analyse data from nearly two million people with chronic conditions to detect changes related to diabetes, hypertension, COPD, heart failure or cognitive decline. [7] This is where an anthropological reading matters. In Foucauldian terms, the system does not simply treat individual patients. It turns an entire population into a permanent field of observation, classification and anticipation. Health ceases to be only a clinical relationship and also becomes a form of governmentality. Knowing a population statistically makes it possible to intervene before the event one wants to prevent has even occurred. [27]
None of this is necessarily problematic. It may anticipate illness, improve diagnoses and free up clinical time. That is precisely why it deserves politics. What data feed these systems? Who can reuse them? How is a false positive measured? Do they work equally well across different social groups? Can a person know why a model has classified them as a high-risk patient? Who audits those who develop the system? Who is accountable when the algorithm is wrong? Madrid’s main annual political debate barely went there.
When the algorithm advises what to study
The new Virtual Education Card will use artificial intelligence to recommend schools according to family and academic profiles. Later, it will analyse each student’s strengths and difficulties and suggest possible pathways, including a decision as consequential as whether to continue with Bachillerato or move into vocational education and training. [8]
The question is not whether an algorithm can detect useful patterns. It can. The question is what kind of society it learns from those patterns.
A system trained on past trajectories can learn which students succeeded. It can also learn the inequalities that produced those same trajectories. Neighbourhood, income, school of origin or apparently neutral correlations can end up acting as proxies for social differences.
We are not claiming that this will happen. We are asking how we will know that it does not. Will a family be able to know why a particular pathway is recommended? Will they be able to challenge it? How much weight will teachers’ judgement carry? Will anyone test whether recommendations change systematically according to income, gender, disability or origin?
Madrid is designing the tool. New York has just taken a different decision.
On 2 September, the largest public-school system in the United States introduced a one-year moratorium on generative AI aimed at pupils from 2-K through eighth grade, affecting nearly 600,000 students. Companion chatbots are banned across all grades. At upper-secondary level, only limited pilots under direct teacher supervision will be allowed. [9]
All upper-secondary students will also have to complete two annual AI-literacy modules covering how the technology works, privacy, bias, risks, ethics and employment consequences. New York will review school technology tools against provider transparency, security, pedagogical evidence and educational impact. More than 38 previously authorised AI components will be disabled because they do not meet the new criteria. [10]
This is not about portraying Madrid as pro-AI and New York as anti-AI. New York is doing something more interesting. It has turned the introduction of technology in schools into a public decision that must be justified. Nor does the comparison always have to look northwards.
In Chile, the bill regulating artificial-intelligence systems is in its second constitutional stage. In July 2026, the Senate’s Future Challenges Committee was still examining it with participation from the Ministry of Science. Chile’s AI policy explicitly places people’s well-being, human rights, safety, inclusion and sustainable development at its centre. Its action plan adds data governance, algorithmic transparency, bias measurement and impact assessment. [28][29]
Chile does not offer a definitive answer, and its institutional capacities differ from Madrid’s. The important point is different. In the Global South too, artificial intelligence is being turned into a parliamentary question about rights, risk and development models, not merely a tool to be deployed.
An administration on a European scale
Madrid’s scale makes this absence even more striking. The Community’s non-financial budget for 2026 reaches €30.6636 billion. [11] Slovenia plans €17.7 billion in state expenditure for 2026 and Lithuania €27.5 billion. [12][13] The competences are not equivalent. A state funds defence, foreign affairs and other functions that Madrid does not assume. The comparison is useful only for visualising institutional scale.
Madrid’s economy reached €336.489 billion in GDP in 2025. [14] For scale, Finland produced around €281 billion in 2025, Czechia around €347 billion and Denmark around €410 billion. [15] Madrid’s economy is therefore larger than Finland’s and close to Czechia’s. Nor are we dealing with a small peripheral administration trying to catch up with the digital revolution. The Community has resources, universities, hospitals, technology companies, a dedicated Regional Ministry for Digitalisation and a sophisticated technical structure. That is precisely why it is reasonable to expect political deliberation equal to its technological capacity.
Hours of politics and an AI debate that arrives at the end
The main interventions from Más Madrid, PSOE and Vox focused on housing, healthcare, corruption, wildfires, inequality, taxation and immigration. These are central issues. What is significant is that almost none of the algorithmic decisions announced by the Government were questioned.
The PSOE provided one exception when it denounced the so-called “Quirón algorithm”, which it said was diverting appointments towards healthcare centres managed by the company. Technology appeared as a possible symbol of opacity, but it did not open a broader discussion about algorithmic governance. [16]
Here we use algorithm in the broad sense in which it appeared in the debate. The material reviewed does not allow us to establish whether this is machine learning, a rules engine or simply automated routing logic. That is precisely why the relevant question is not whether to call it AI, but what rules it uses, who decided them and whether they can be audited. The broader discussion had to wait until almost the end.
During the defence of the resolutions, Emilio Delgado of Más Madrid spoke about the risks of unregulated artificial intelligence for hospitals, water-treatment plants, airports and internet infrastructure. He then did something no one had done during the major interventions. He connected the Madrid debate to a warning that was unfolding almost in real time. [16]
On 8 September, Jacob Coxon had announced his resignation from Anthropic and warned that the race towards systems capable of improving themselves was advancing without sufficient guarantees of control. Evan Hubinger, Anthropic’s head of Alignment Science, publicly endorsed the seriousness of that warning and put his own estimate of the risk of human extinction from AI at above 10% over the next decade. He also acknowledged that Anthropic still does not have a plan for solving the alignment problem of a possible superintelligence. [30][31]
Delgado also cited Geoffrey Hinton as another voice warning of catastrophic risks. Hinton received the 2024 Nobel Prize in Physics, together with John Hopfield, for foundational work on machine learning with neural networks. [32] There is no need to share the existential-risk estimates of Coxon, Hubinger or Hinton to notice the anomaly. While people working inside companies building some of the most advanced models were publicly discussing whether such systems might escape human control, a single voice in the Assembly introduced that conversation in Madrid, at the end of the session, and almost no one picked it up.
For the first time, the question was no longer only what AI can do for the Administration. It became what power AI creates and who controls it. The conversation was possible. It simply arrived when the debate was almost over.
When Parliament starts to resemble the feed
There is also a question of form. An Assembly should be one of the places where politics can slow down. A figure can be challenged. A hypothesis can be distinguished from a fact. A public policy can be examined without the immediate pressure of the click. Much of the debate had a different texture.
The penthouse. The partner. The brother. The trips. The properties. Morocco. Sánchez. Ministers. The accusation and the response to the accusation.
That does not make the issues discussed irrelevant. Housing, corruption, healthcare and migration are central questions. What matters is that this form of conversation looks remarkably similar to the kind of content that platform recommendation systems know how to reward.
An experiment published in Nature in 2026 found that feeds ranked by engagement, compared with chronological ordering, amplified intergroup political content that was moralised, emotional and toxic. Another experimental study of X found effects on political exposure and some opinions. [17][18]
We cannot conclude that platforms caused the form of the Madrid debate. We can ask what happens to a deliberative institution when those speaking know that every sentence is simultaneously competing to become a clip. The Assembly should be one of the places where politics becomes de-algorithmised. For many hours, almost the opposite happened.
From data to dossier
Ayuso takes this mechanism particularly far in her response. She lists homes, square metres, garages, trips, renovations and properties attributed to members of Spain’s national government and people around them. We do not need to imagine a secret database. Much of this information can be reconstructed from asset declarations, public registers, budgets, official agendas and the press. What matters is how it is then combined. Documented facts, media reports, interpretations and suspicions appear within the same argumentative sequence.
At one point Ayuso states that “Morocco spied on Pedro Sánchez’s phone” and presents it as “another fact”. [16]
We know that the Prime Minister’s phone was infected with Pegasus. We do not know judicially who ordered it. Spain’s National Court again provisionally closed the investigation in January 2026 because Israel did not cooperate with letters rogatory. That lack of cooperation prevented responsibility from being attributed to any specific actor. [19] Morocco remains a relevant hypothesis. A hypothesis and a fact are not the same thing.
In an ecosystem where platforms, search engines and generative models compress millions of fragments of information, preserving the difference between data, indication, hypothesis and interpretation has itself become part of democratic infrastructure.
The minister, Palantir and Israel
The paradox becomes sharper when we look beneath the speech. Miguel López-Valverde, Madrid’s Regional Minister for Digitalisation, does not come primarily from politics. He is a telecommunications engineer and spent more than two decades in the technology sector. In 2021, from Inetum, he managed projects for Quirónsalud and Quirón Prevención. In 2022 he became head of Information Systems and Digital Health at SERMAS. In 2023 he became a regional minister. [20] His background helps explain why healthcare now sits at the centre of the region’s algorithmic deployment, and it also reveals another infrastructure that barely appeared in the Debate. The Community of Madrid has formal contacts with Palantir Technologies.
Civil Protection met the company on 19 November 2024 and 17 January 2025 to examine disaster-management tools. On 25 November 2025, López-Valverde himself met Palantir representatives to explore possible forms of collaboration with the Administration. The Transparency Portal also records a fourth meeting on 29 April 2026, convened by Palantir Technologies and assigned to the Deputy Ministry for Justice and Victims. [21][22][23][24]
In the public sources reviewed for this Radar, we have found no evidence that Palantir is deployed in production within the Community or that Madrid’s PP uses its systems for political intelligence.
But the framework developed by Nick Couldry and Ulises A. Mejias helps clarify what is at stake if such a relationship advances. Their concept of data colonialism describes how social life can become extractable raw material for private processing infrastructures. Applied here with caution, what matters is not only where public data are physically or legally held, but who has the technical capacity to cross-reference them, interpret them and turn them into operational knowledge. [33] A public administration may formally retain ownership of the data while still shifting part of its capacity to know towards a provider’s architecture. Four meetings do not demonstrate deployment, but they do justify asking what happened next.
The same applies to Israel.
In January, López-Valverde led an institutional mission that included the Israel National Cyber Directorate, Check Point, Pentera and Rafael Advanced Defense Systems. The Community examined the Cyber Dome model, based on AI, Big Data and continuous monitoring to protect critical infrastructure. [25][26]
In the public sources reviewed, we have likewise found no relationship between the Community of Madrid or Madrid’s PP and NSO Group or Pegasus. Palantir is not Pegasus, and cybersecurity is not espionage. That is precisely why the categories should not be blurred. But healthcare AI, large-scale data analysis, educational profiling, administrative agents, Palantir and cyber-defence models are already part of the real environment in which Madrid’s Administration is evolving. Almost none of it occupied the centre of its main political debate.
A cutting-edge region needs a cutting-edge democracy
Madrid has grounds to regard itself as a technologically advanced region. It has resources, universities, hospitals, global companies, data centres, talent and a population that rapidly adopts new tools. But being at the cutting edge is not only about installing technology earlier. It is also about developing the institutions capable of governing it earlier.
Who controls the models. Who audits them. What rights a person has when faced with an automated classification. What data are used. Which companies gain access to them. What criteria determine an educational or medical recommendation. Which jobs will change. What material infrastructure artificial intelligence requires. What power becomes concentrated around it. None of this requires being against AI. It requires taking AI seriously as a political matter.
While those questions were barely making their way into the Assembly, the systems were already entering hospitals, case files, schools and public services. Perhaps that is the real delay. Artificial intelligence has already entered Madrid. Politics is still catching up.
References
-
Comunidad de Madrid. La Comunidad de Madrid aprueba su primera Ley de Administración Digital e Inteligencia Artificial. 8 September 2026.
https://www.comunidad.madrid/noticias/2026/09/08/comunidad-madrid-aprueba-su-primera-ley-administracion-digital-e-inteligencia-artificial -
Instituto Nacional de Estadística. Encuesta sobre Equipamiento y Uso de Tecnologías de la Información y Comunicación en los Hogares. Año 2025.
https://www.ine.es/dyngs/Prensa/es/TICH2025.htm -
Instituto Nacional de Estadística. Encuesta sobre el uso de TIC y del comercio electrónico en las empresas. Año 2024 - Primer trimestre 2025. Data by autonomous community.
https://www.ine.es/dyngs/Prensa/ETICCE20241T2025.pdf -
Comunidad de Madrid. La Comunidad de Madrid cuenta ya con 220 casos de uso desarrollados con Inteligencia Artificial, de los cuales 96 ya están implantados. 11 February 2026.
https://www.comunidad.madrid/noticias/2026/02/11/comunidad-madrid-cuenta-ya-220-casos-uso-desarrollados-inteligencia-artificial-cuales-96-ya-estan-implantados -
Transcript of the State of the Region Debate in the Community of Madrid, 10 September 2026. Working material supplied for this Radar.
-
Comunidad de Madrid. Plan Estratégico de Salud Digital 2026-2028.
https://www.comunidad.madrid/digitalizacion/plan-estrategico-salud-digital-2026-2028 -
Comunidad de Madrid. Proyectos de inteligencia artificial para anticipar enfermedades crónicas y personalizar la atención sanitaria. 10 September 2026.
https://www.comunidad.madrid/noticias/2026/09/10/comunidad-madrid-desplegara-proyectos-inteligencia-artificial-anticipar-enfermedades-cronicas-personalizar-atencion-sanitaria -
Comunidad de Madrid. Tarjeta Virtual Educativa. 10 September 2026.
https://www.comunidad.madrid/noticias/2026/09/10/diaz-ayuso-anuncia-tarjeta-virtual-educativa-tener-solo-clic-toda-vida-academica-alumnos-madrilenos -
City of New York. Mayor Mamdani and Chancellor Samuels put students first with new AI policy. 2 September 2026.
https://www.nyc.gov/mayors-office/news/2026/09/mayor-mamdani-and-chancellor-samuels-put-students-first-with-nat -
City of New York. Transcript: Mayor Mamdani holds press conference on education and AI. September 2026.
https://www.nyc.gov/mayors-office/news/2026/09/transcript--mayor-mamdani-holds-press-conference-to-make-educati -
Comunidad de Madrid. Presupuestos Generales de la Comunidad de Madrid 2026.
https://www.comunidad.madrid/presupuestos/presupuestos-generales-comunidad-madrid-2026 -
Government of Slovenia. State budgets for 2026 and 2027.
https://www.gov.si/en/news/2025-09-25-170th-regular-session-of-the-government-of-the-republic-of-slovenia/ -
Ministry of Finance of Lithuania. Seimas approves the State budget for 2026-2028.
https://finmin.lrv.lt/en/news/seimas-approves-the-state-budget-for-2026-2028-VkAc/ -
Comunidad de Madrid. La economía madrileña cerró 2025 con un crecimiento del 3 %. 12 March 2026.
https://www.comunidad.madrid/noticias/2026/03/12/diaz-ayuso-anuncia-economia-madrilena-cerro-2025-crecimiento-3-dos-decimas-encima-media-espana -
Statistisches Bundesamt (Destatis). Germany in the European Union: Key figures 2026. GDP at market prices for Finland, Czechia and Denmark in 2025.
https://www.destatis.de/Europa/EN/Topic/Key-indicators/Uebersicht.html -
Transcript of the second day of the State of the Region Debate, 11 September 2026. Contributions from Más Madrid, PSOE, Vox and PP, government replies and resolutions. Working material supplied for this Radar.
-
Nature. Experimental study of recommendation systems, engagement and political conversation. 2026.
https://www.nature.com/articles/s41586-026-10536-1 -
Nature. Experimental study of X’s algorithmic feed. 2026.
https://www.nature.com/articles/s41586-026-10098-2 -
Poder Judicial. El juez de la Audiencia Nacional archiva por segunda vez el caso Pegasus ante la obstrucción de Israel a cooperar con la investigación. 22 January 2026.
https://www.poderjudicial.es/cgpj/en/Judiciary/National-High-Court/Judicial-News/El-juez-de-la-Audiencia-Nacional-archiva-por-segunda-vez-el--caso-Pegasus--ante-la-obstruccion-de-Israel-a-cooperar-con-la-investigacion -
Comunidad de Madrid. Miguel López-Valverde Argüeso. Perfil y trayectoria.
https://www.comunidad.madrid/transparencia/persona/miguel-lopez-valverde-argueso-0 -
Comunidad de Madrid. Reunión con la empresa Palantir. 19 November 2024.
https://www.comunidad.madrid/transparencia/agenda/reunion-empresa-palantir -
Comunidad de Madrid. Reunión con la empresa Palantir. 17 January 2025.
https://www.comunidad.madrid/transparencia/agenda/reunion-empresa-palantir-0 -
Comunidad de Madrid. López-Valverde se reúne con Palantir para analizar vías de colaboración con la Administración pública. 25 November 2025.
https://www.comunidad.madrid/agenda-gobierno/2025/11/24/lopez-valverde-reune-empresa-estadounidense-palantir-especializada-analisis-datos -
Comunidad de Madrid. Reunión con Palantir Technologies. 29 April 2026.
https://www.comunidad.madrid/transparencia/agenda/reunion-palantir-technologies -
Comunidad de Madrid. La Comunidad de Madrid y la Agencia Nacional de Ciberseguridad israelí intercambian conocimientos y experiencias. 25 January 2026.
https://www.comunidad.madrid/noticias/2026/01/25/comunidad-madrid-agencia-nacional-ciberseguridad-israeli-reunen-intercambiar-conocimientos-experiencias-materia -
Comunidad de Madrid. López-Valverde visita Rafael Advanced Defense Systems. 22 January 2026.
https://www.comunidad.madrid/agenda-gobierno/2026/01/22/lopez-valverde-visita-karmiel-israel-empresa-rafael-advanced-defense-sytems-especializada-tecnologia-defensa -
Foucault, Michel. Security, Territory, Population: Lectures at the Collège de France, 1977-1978. Palgrave Macmillan.
https://link.springer.com/book/10.1057/9780230245075 -
Senado de la República de Chile. Comisión de Desafíos del Futuro, Ciencia, Tecnología e Innovación. Session of 13 July 2026 on the regulation of artificial-intelligence systems.
https://www.senado.cl/actividad-legislativa/comisiones/941/22741 -
Ministerio de Ciencia, Tecnología, Conocimiento e Innovación de Chile. Política Nacional de Inteligencia Artificial.
https://www.minciencia.gob.cl/areas/inteligencia-artificial/politica-nacional-de-inteligencia-artificial/ -
Axios. Anthropic whistleblower gave up his equity to leave the company. 9 September 2026.
https://www.axios.com/2026/09/09/anthropic-researcher-ai-warning-interview -
Forbes. Anthropic Researcher Warns There’s ‘>10% Chance’ AI Could ‘Kill All Humans’ By Next Decade. 9 September 2026.
https://www.forbes.com/sites/siladityaray/2026/09/09/anthropic-alignment-lead-warns-ai-could-kill-all-humans-as-researcher-quits/ -
Nobel Prize. The Nobel Prize in Physics 2024. 8 October 2024.
https://www.nobelprize.org/prizes/physics/2024/press-release/ -
Couldry, Nick and Ulises A. Mejias. The Costs of Connection: How Data Is Colonizing Human Life and Appropriating It for Capitalism. Stanford University Press, 2019.
https://www.sup.org/books/title/?id=28816