The OECD.AI Policy Navigator

Our policy navigator is a living repository from more than 80 jurisdictions and organisations. Use the filters to browse initiatives and find what you are looking for.

PresidencIA: Generative AI Adoption Programme for the Spanish Prime Minister’s Office


-
Added by:   OECD analyst
Added on:   17 Jul 2026
Updated by:   OECD analyst
Updated on:   17 Jul 2026

Leveraging on 3 pillars (Community, Applications and Infrastructure), PresidencIA builds AI literacy among staff through workshops, community and best practices, develops custom AI tools for processes in the PM’s Office, and pursues on-premises AI infrastructure for tech sovereignty. It also works to position Spain as a partner for leading AI labs on safety and alignment. The programme addresses the gap between AI’s potential and practical government capacity.

Name in original language

PresidencIA: Programa de Adopción de IA Generativa para el Gabinete de Presidencia del Gobierno de España

Initiative overview

PresidencIA is a comprehensive programme designed to accelerate the responsible adoption of generative AI within the Spanish Prime Minister’s Office (Gabinete de Presidencia). While AI offers significant potential to improve public‑sector decision‑making and service delivery, many administrations lack the internal capacity, tools and infrastructure needed to adopt it safely. PresidencIA responds by combining skills development, purpose‑built applications, sovereign infrastructure and international engagement into a single, integrated approach.

The programme’s first pillar focuses on community and capacity building. Through hands‑on workshops such as Whispering to AIs, Prototyping with AI and Researching with AI, several hundred public servants have developed practical skills in prompt engineering and responsible AI use. These activities prioritise experimentation and peer learning over theoretical instruction, supported by an active community of practice where staff share experiences and best practices. 

The second pillar delivers AI applications tailored to the Prime Minister’s Office. These include ServetIA, a research platform that produces source‑verified analytical outputs to support policy work; CiudadanIA, which automates the summarisation of citizen correspondence addressed to the President, reducing processing time and improving response quality; and GlobalAlert, which provides real‑time geopolitical monitoring and threat analysis. The programme also developed PromptAventura, a gamified AI literacy tool now validated and transferred to the government intranet. In addition, PresidencIA produced open‑source components, including an LLM evaluation workbench used to test prompts and models at scale through LLM‑as‑a‑judge methodologies.

The infrastructure pillar addresses AI sovereignty and secure deployment. PresidencIA has tested open‑weight language models for on‑premises use, enabling sensitive data processing without reliance on external cloud providers. In collaboration with the government IT unit (U.TIC), the programme is assessing the feasibility of a dedicated internal AI compute cluster and fine‑tuning models for specific government use cases. Alongside domestic infrastructure, PresidencIA also pursues international engagement, positioning Spain as a strategic partner for leading AI labs and fostering collaboration on alignment, safety research and responsible deployment of frontier AI systems in government contexts.

Other relevant details

PresidencIA is designed to be scalable and replicable. Its modular structure—combining skills development, practical tools, sovereign infrastructure and global engagement—provides a transferable model for other public institutions seeking to adopt AI responsibly. Concrete results are already visible. Hundreds of staff are trained and actively participating in the AI community. ServetIA is operational for policy research, CiudadanIA is processing large volumes of citizen correspondence, GlobalAlert supports situational awareness, and PromptAventura is used for gamified learning. The programme has also delivered an on‑premise LLM evaluation framework and initiated dialogue with leading AI labs. Impact is monitored through participation levels, tool adoption metrics and user feedback. Expected benefits include reduced research time, faster citizen response handling, improved crisis awareness and a reusable model for wider government adoption. Implementation revealed several challenges: Operating in a high‑security environment required careful coordination on cybersecurity and risk management. Staff arrived with very different skill levels, and balancing innovation speed with institutional risk tolerance required incremental deployment. Quality assurance for smaller, quantised language models was particularly important in sensitive applications such as CiudadanIA, where early outputs hallucinated information not present in original citizen letters. These issues were addressed through supervised fine‑tuning, reinforcement learning techniques, systematic prompt optimisation based on expert feedback and large‑scale evaluation using LLM‑as‑a‑judge methods. Key lessons highlight that practical, hands‑on training is far more effective than theoretical guidance, and that trust is built through visible, well‑scoped successes before scaling. Early engagement with IT security teams reduced friction, while international partnerships required demonstrating both technical credibility and strong political commitment. Overall, sustained success depends on senior leadership support and embedding AI initiatives within existing institutional structures rather than treating them as isolated experiments.