· The initiative, which has counted on the collaboration of Amazon AWS and Multiverse Computing, has been designed to produce improvements in the 40B Instructed model, with the aim of moving towards a safer, more egalitarian and more inclusive artificial intelligence.
· ALIA is the family of artificial intelligence models promoted by the Government of Spain, trained in Spanish and co-official languages, which is already being applied by the business fabric and the Public Administrations.
The Government of Spain continues to make progress in the development of a more secure, egalitarian and inclusive artificial intelligence, with the involvement of citizens. This has been demonstrated today in the final of the 2nd ALIA Challenge, held at Talent Arena 2026, a co-location event at Mobile World Congress Barcelona, with the support of the Ministry for Digital Transformation and the Public Service. The challenge, promoted by the Secretary of State for Digitalization and Artificial Intelligence, has managed to summon more than 1,000 people to collaborate in the improvement of the public family of language models developed by the Barcelona Supercomputing Center-National Center for Supercomputing.
The challenge, which has benefited from the collaboration of Amazon AWS and Multiverse Computing, has been designed to produce tangible improvements in the behavior of the 40B Instruido model that, when faced with problematic requests, must identify risks, set limits and offer useful alternatives without providing improper uses. In addition, the purpose has been marked for the model to use a more equitable and inclusive language, avoiding the reproduction of stereotypes, and to bring AI closer to citizens. Thus, it complements the objectives of the first edition, held in December 2025, focused on proposing solutions to achieve a more sustainable and responsible AI with energy resources, within the framework of the National Plan of Green Algorithms.
Participants from 48 provinces and 18 countries
The 2nd ALIA Challenge has been developed in two phases. The first, online, free registration and directed to people without the need for technical knowledge, was attended by 974 people from 48 Spanish provinces and 18 countries (Peru, Argentina, Chile, Colombia, France, Italy, or the United Kingdom, among others). Of the total, 64.8% were men and 31.9% women, a figure higher than 20% of women engaged in engineering in Spain.
Enrollees tested the model by selecting one of the proposed alignment categories (defamation, crime, hate, legal/medical/financial advice or gender bias) and, based on a conversation with the machine, the participant validated the responses provided by ALIA or, on the contrary, offered a correct response in case the system had incurred errors (such as responding with gender bias). In total, up to 29,404 attempts were made for ALIA to provide erroneous responses and 1,780 proposals for responses following ethical and safety principles have been made to the model.
After the end of the online phase, all the interactions with the system have been collected in a data set that has been worked on in the second phase, of a face-to-face nature, consisting of a hackathon developed yesterday and today in Talent Arena. The 85 participants, organized by teams, have faced two main challenges: validating a data set with human judgment (“educating” AI); and designing and testing an AI “judge” with the validated data set, operating according to human preferences, assessing quality and correctness of responses according to established ethical and safety principles.
This face-to-face phase has culminated in the award ceremony, with a total value of 9,000 euros, according to different established criteria. The winners were Team 12 with the 1st Prize (Manuel Antón Herrerías, Azucena Vázquez, Katherina Akhmadeeva, Igor Soroka and Iván Boris), Team 6 with the 2nd Prize (Abraham Gil Ramírez, Manel Arribas Ibar, Paula Alcaide Barbero, Pol Serrano Franquesa and Pedro Latasa Fernández-Layos) and Team 19 with the 3rd Prize (Guillem Farriols, Ignrisasi, Gómez). They have received the recognitions from the Secretary of State for Digitalization and Artificial Intelligence, María González Veracruz; Cristian Cantón, associate director of the Barcelona Supercomputing Center-National Center for Supercomputing; and María Jesús Martín, director of the Governance and Planning Division of the General Directorate of Artificial Intelligence of the Secretariat of State.
In her speech, María González Veracruz thanked the citizens for their participation so that “with your knowledge, with your effort, you are contributing to a country project, improving and aligning ALIA, an open, safe, egalitarian and unbiased public artificial intelligence, which is already being used by the business fabric and the Public Administration to improve people’s lives.” In addition, he has highlighted that “with this collective work we are laying the foundations for a community of artificial intelligence developers in Spain”. And finally, he highlighted “the success of the Talent Arena national tour that we have developed together and that has been highlighted in Madrid, Murcia, Valencia and Zaragoza, in collaboration with several public universities”.
Amazon AWS and Multiverse Computing, Challenge Partners
"At AWS, we are very proud to be able to collaborate with this pioneering initiative of the Ministry for Digital Transformation and Public Service along with Multiverse Computing. The ALIA Challenge represents a milestone in the development of sovereign and responsible artificial intelligence in Spain, and this collaboration demonstrates precisely our commitment to innovation in the public sector and our ability to provide the necessary infrastructure that allows more than 1,000 people to contribute to ethically improve the ALIA-40B model,” said Óscar Sanz, Head of Government for the Public Sector at AWS Spain.
“For Multiverse Computing, the ALIA Challenge is a key step towards the development of sovereign and European artificial intelligence. In this context, we have supported the development of the ALIA-40B model by producing data sets in Spanish and co-official languages, with the aim of obtaining a stable and multilingual AI model. This collaboration promoted by the Ministry for Digital Transformation and the Public Service also contributes to strengthening and developing the artificial intelligence ecosystem in Spain,” said Iraia Ibarzabal, Chief Growth Officer of Multiverse Computing.
The ALIA project: the sovereign AI of the Spanish Government
ALIA is a strategic initiative of the Secretary of State for Digitalization and Artificial Intelligence of the Ministry for Digital Transformation and Public Function, developed by the Barcelona Supercomputing Center-National Center for Supercomputing. Among its objectives are to create open foundational models such as ALIA 40B (40 billion parameters), reuse through public infrastructure and ensure ethical standards aligned with the European AI Regulation and supervised by AESIA.
For the business fabric, which is already applying the models to real artificial intelligence solutions, as announced by Minister Óscar López in the framework of the Mobile World Congress, ALIA opens up opportunities by reducing barriers to access and stimulating responsible innovation. ALIA is expected to consolidate the position of Spain as a European reference in the deployment of responsible and auditable language models applied to public services, strengthening collaboration between administrations, companies and research centers.
