Spain’s Sánchez launches Plan IA360 — “self-regulation does not work”

Spanish Prime Minister Pedro Sánchez presented Plan IA360 at Moncloa Palace — a 12-month government roadmap for responsible AI — and said industry self-regulation fails when a few firms control the most advanced systems.

POLITICS desk — A major European capital answered this week’s argument over slowing AI versus letting labs police themselves, with a concrete 12-month plan and an explicit no to industry self-regulation.

Sánchez said self-regulation does not work. La Moncloa quotes him: history shows self-regulation fails, and the incentives of the few who control these frontier industries will never line up with the public interest. Frontier here means the most advanced AI systems, the ones at the edge of what labs can build. Reuters’ same-day lead is the same point in the wire’s words: artificial intelligence cannot be self-regulated by those who control the technology. He also warned, in the official readout, that an unregulated, opaque AI “stampede” in a few hands is a recipe for disaster, and that AI without the control of collective human intelligence “is not intelligence: it is chaos and stupidity.” File those lines as his. This desk did not interview him.

The plan has four goals, as La Moncloa lays them out. First, strengthen Spain’s tech capacity. Named pieces: an AI gigafactory — a very large computing plant for AI, still a project in the plan, not a building this desk has seen — and new AI models built with the Barcelona Supercomputing Center for climate, energy, and health uses. Data centers in the plan are supposed to meet high energy and environmental standards and keep Spain in control of the data stored there. Reuters adds his line that the technology should not be developed “behind citizens’ backs,” and that people should see it as an opportunity rather than a tool for using up scarce local resources. File the gigafactory, the Barcelona models, and the data-center standards as the plan’s. Do not invent a budget in euros, a construction date, or a claim the gigafactory is already running.

Second, turn fast public uptake of AI into an economic engine and a training ground. The tool is an “AI voucher” — in Spanish, Bono Inteligencia Artificial — for small and medium-sized firms and for self-employed people. The aim, as Sánchez stated it, is that by 2030 more than half of Spanish companies have built AI into their work. That is a majority, not a claim it has already happened. To get there, the government says it will update the secondary-school curriculum and vocational training. Vocational training is FP, formación profesional, Spain’s job-skills track. The stated purpose: students learn to use AI and live with it, and the rollout does not damage how they learn. File the voucher, the 2030 aim, and the curriculum update as La Moncloa’s. This desk did not see a voucher form or a new textbook.

Third, governance. Sánchez called the security side an absolute priority, on two tracks: a stronger cybersecurity shield, and tighter supervision of the most aggressive frontier models. Reuters says he would reinforce defensive cybersecurity measures against what he called “aggressive” frontier models. File that pair as the plan’s. This is an oversight promise, not a new criminal statute published today.

Fourth, a “Social Contract for AI” — Contrato Social por la Inteligencia Artificial — a national pact among employers, workers, unions, political parties, and public authorities. The point, as he framed it, is to give people security when AI changes jobs and daily life. La Moncloa says the government will convene social partners next month to start that agreement. Social partners here means the unions and employer groups that sit down with the government. The Next Web’s same-day paraphrase carries the same next-month meeting. File the pact and the meeting as announced, not as a signed contract. No text of the pact was published.

Sánchez also argued companies should be legally and financially liable for harms from their AI. La Moncloa names three examples: deepfakes — fake but realistic video, audio, or images of real people — harm to the mental health of children and young people, and theft of intellectual property, meaning someone else’s creative or industrial work. He framed that liability as part of a global push with other countries through bodies that already exist, including the EU AI Office and a United Nations scientific panel. File the liability argument as his. Do not invent a new damages law, a fine schedule, or a court case.

Spain’s cited record, still the official readout: AESIA, the Agencia Española de Supervisión de la Inteligencia Artificial. La Moncloa says Spain was the first European country to create a dedicated AI supervisory agency. It also says the EU AI Act — the European Union’s comprehensive AI rulebook — was agreed while Spain held the EU’s rotating presidency. File AESIA and that presidency agreement as the government’s account. Adoption percentages and a Stanford governance ranking in the same readout stay in Sources. This desk did not audit them.

Do not read this card as Spain’s national AI bill becoming law. The Next Web, a same-day paraphrase, says that national AI law is still a bill. The cabinet sent it to parliament in May, and it is not expected to pass before the end of 2027. Until then, The Next Web says, binding rules come from EU regulations, not from a new Madrid statute. Plan IA360 is the 12-month roadmap announced today. It is not that bill, and it is not a claim the bill passed.

Do not invent names for the “techno-oligarchs.” La Moncloa says Sánchez criticized magnates who still call regulation the enemy of innovation, who think they stand above the state, and who think that if rules exist they should write them. He said they are wrong, and that nobody is above the public interest. Reuters calls him an outspoken critic of “tech oligarchs.” He did not name the people in the official readout. Do not attach this speech to a named lab-CEO quote he did not give.

Plain English for the rest of the card: Plan IA360 = Spain’s 12-month government roadmap for responsible AI, announced 21 Sep 2026. IA = inteligencia artificial, Spanish for AI. Moncloa = the prime minister’s palace and the name of the official government site. frontier model = one of the most advanced AI systems. gigafactory = a very large AI computing plant — named as a project, not opened today. AI voucher / Bono Inteligencia Artificial = a planned aid for small firms and the self-employed to adopt AI. FP / formación profesional = Spain’s vocational training track. Social Contract for AI = a proposed national pact on jobs and daily life, not a signed text. social partners = unions and employer groups. AESIA = Spain’s AI supervisory agency. EU AI Act = the European Union’s AI rulebook, already agreed, which this desk is not reprinting as a new Spanish law. deepfake = a realistic fake image, video, or audio of a real person. This filing is the Monday roadmap and the self-regulation line. It is not a passed national bill and not a euro budget.

PRIMARY here: La Moncloa’s 21 Sep 2026 official readout of Pedro Sánchez’s closing remarks at the Plan IA360 event — Tier A PRIMARY government source, the original record. Reuters’ same-day Madrid dispatch is independent Tier B confirm of the self-regulation line, the 12-month roadmap, the proposed AI gigafactory, the Barcelona Supercomputing Center models for climate, health, and energy, the data-center standards, the cybersecurity reinforcement against “aggressive” frontier models, and the not-behind-citizens’-backs line. The Next Web is a same-day paraphrase — discovery corroboration of the next-month social-contract convening and of the national AI law still being a bill, not a substitute primary and not a source for a euro budget. The 12-month roadmap, the self-regulation / public-interest line, the stampede warning, the four goals, the gigafactory, the Barcelona models, the data-center standards, the AI voucher, the 2030 more-than-half aim, the secondary and vocational curriculum update, the cybersecurity-plus-frontier-supervision pair, the Social Contract for AI, the next-month convening, the legal-and-financial liability argument with deepfakes / youth mental health / intellectual-property examples, AESIA as the government’s account of the first dedicated AI agency a European country created, and the EU AI Act agreed under Spain’s EU presidency are La Moncloa-attributed. The cannot-be-self-regulated lead, the “aggressive” frontier-model line, and the citizens’-backs quote are Reuters-attributed. The national bill still not law, the May referral to parliament, and the not-expected-before-end-2027 line are The Next Web-attributed. NOT claimed: a euro budget, that the AI gigafactory is already built, that Spain’s national AI bill passed, a named techno-oligarch he refused to name, a signed social contract, a new damages statute, that this desk sat in the room, a stock tip, or investment advice. Distinct from the already-filed trump-ai-force-czar, bessent-china-ai-dialogue, and trump-rejects-ai-guardrails-amodei.