SaturdaysAI

A shared baseline for AI safety and evaluation: agreement on minimum international standards for testing, auditing, and monitoring advanced AI systems, including open benchmarks and red-teaming practices. Practical cooperation mechanisms: creation of working groups or task forces (e.g., on safety, capacity-building, and open ecosystems) with clear deliverables before the 2027 meeting. Global capacity-building commitments: funding, tools, and partnerships to ensure that emerging economies can meaningfully participate in AI development and governance, not just regulation. Interoperability across regulatory regimes: initial alignment on key definitions, risk tiers, and compliance approaches to avoid fragmentation while respecting regional differences. Support for open and inclusive AI ecosystems: recognition of the role of open-source models, open data, and shared infrastructure as public goods that democratize innovation. Multi-stakeholder integration: formal inclusion of practitioners (educators, builders, startups, civil society) alongside governments to ground discussions in real deployment contexts.

• AI capacity-building
• Open-source software, open data and open AI models
• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI

Safe, secure, and trustworthy AI is foundational. Without shared approaches to evaluation and safety, global deployment risks outpacing our ability to manage systemic failures or misuse. AI capacity-building is critical to avoid a two-speed world. Most talent and innovation potential is globally distributed, but access to compute, training, and practical education is not. Initiatives like community-driven education and hands-on programs have shown that capability can scale quickly if supported. Social, economic, and cultural implications must be addressed early. AI is already reshaping labor markets, education, and access to knowledge. Governance must integrate these dimensions, not treat them as secondary effects. Open-source, open data, and open models are key enablers of inclusion and transparency. Open ecosystems reduce dependency on a small number of actors, accelerate learning, and allow local adaptation, particularly important for linguistic and cultural diversity. I consider this a pragmatic approach: ensure safety, expand participation, understand impact, and keep the ecosystem open enough to remain innovative and equitable.

Verification and trust infrastructure: As AI-generated content scales, verifying authenticity (text, media, agents) becomes a bottleneck. Governance should address provenance standards, watermarking limits, and economic models for verification. Human-AI collaboration and augmentation: Most real-world impact comes from hybrid systems. Policies should consider how AI augments human decision-making, not just replaces it, including implications for education and professional training. Compute and infrastructure access: Access to compute is a structural determinant of who can build and benefit from AI. This includes cloud credits, shared infrastructure, and potential public compute initiatives. Agentic AI and autonomy: Emerging systems that act over time (agents) introduce new risks (goal misalignment, cascading actions) and require updated oversight models beyond static evaluations. Data governance beyond privacy: Questions of data ownership, compensation, and collective data rights are becoming central, especially for communities contributing to training data. Geopolitics of AI ecosystems: Fragmentation between regions (e.g., US, China, EU) affects standards, supply chains, and safety coordination. Mechanisms for minimal cooperation are essential.

From a combined Spain–Europe–Spanish-speaking (Ibero-American) perspective, governance gaps in AI are creating both structural risks and unique opportunities. --Challenges-- Fragmentation across regions: The EU AI Act provides a strong regulatory anchor, but alignment with Latin American countries is limited. Diverging standards risk creating barriers for startups, cross-border products, and shared innovation in the Spanish-speaking ecosystem. Capacity and infrastructure gaps: While parts of Europe have access to talent, compute, and funding, many Spanish-speaking countries face constraints in all three. This creates an uneven playing field and limits meaningful participation in AI development. Language and cultural representation: Spanish is globally significant, yet underrepresented in high-quality datasets and models compared to English, affecting performance, inclusivity, and local relevance. Verification and trust deficit: As AI-generated content scales, institutions (media, public sector, education) struggle to verify outputs, increasing risks of misinformation and eroding trust. Dependence on non-local providers: Both Europe and Latin America rely heavily on external AI infrastructure and models, raising concerns about sovereignty, resilience, and long-term competitiveness. --Opportunities-- A shared linguistic and cultural space: The Spanish-speaking world can act as a large, coordinated ecosystem for AI development, including shared datasets, benchmarks, and open models tailored to linguistic diversity. Bridging Europe and Latin America: Spain is uniquely positioned to act as a connector—aligning regulatory approaches, enabling talent exchange, and fostering joint innovation initiatives. Open-source as an equalizer: Open models and data can lower barriers to entry, enabling startups, educators, and public institutions across the region to build and adapt AI solutions locally. Scalable AI education and talent networks: Community-driven initiatives can rapidly expand applied AI skills across both regions, reducing the gap between policy ambition and execution. Public sector and sectoral transformation: AI adoption in government, education, agriculture, and SMEs can drive productivity and inclusion if supported by interoperable standards and accessible tools. Reducing fragmentation, investing in shared capacity, and leveraging the Spanish-speaking ecosystem as a coherent AI space are key to turning governance into impact.

It can turn fragmented governance efforts into coordinated action. Its role is to enable convergence on core elements, such as safety evaluations, risk frameworks, and auditing, while allowing flexibility. It can also catalyze shared public goods (e.g., benchmarks, datasets, testing protocols) that lower barriers to participation, especially for underrepresented regions and languages. By bringing together governments, practitioners, and open communities, it ensures governance is grounded in real-world deployment. This is key to bridging regions like Europe and the Spanish-speaking world. Ultimately, its value is in aligning, operationalizing, and scaling existing efforts into effective global cooperation.

The AI Dialogue should build on existing foundations, such as the UN High-Level Advisory Body on AI, OECD and GPAI work, the G7 Hiroshima Process, the EU AI Act, AI Safety Institutes, and standards bodies like ISO/IEC and IEEE, while connecting open-source ecosystems and capacity-building networks across regions. Its added value is not to duplicate these efforts, but to act as a neutral coordination layer: aligning definitions and evaluation practices, promoting shared public goods like benchmarks and datasets (including for underrepresented languages such as Spanish), and ensuring that emerging economies and practitioner communities are meaningfully included. In doing so, it can reduce fragmentation, bridge regions like Europe and Latin America, and translate existing principles and standards into more coherent, actionable global cooperation.

Governments should share regulatory approaches, pilot interoperable frameworks, and commit to joint initiatives (e.g., safety standards, public-sector deployments). Industry and startups should provide real-world deployment insights, evaluation practices, and open tools where possible. Academia and safety institutes should advance methodologies for testing, auditing, and risk assessment. Civil society and educators should represent societal impacts, inclusion, and capacity-building needs. Open-source communities should contribute shared models, datasets, and benchmarks as global public goods. --Recommended format and structure-- Thematic working groups (e.g., safety, capacity, open ecosystems, emerging risks) with clear deliverables between annual meetings. Hybrid model (top-down + bottom-up): high-level plenaries for alignment, complemented by practitioner-led sessions grounded in real use cases. Action-oriented outputs: each cycle should produce concrete artifacts (e.g., shared benchmarks, policy templates, pilot collaborations), not just statements. Regional bridges: dedicated tracks to connect ecosystems (e.g., Europe-Latin America), leveraging shared language and context. Open participation layer: structured calls for input, public repositories, and transparent reporting to include smaller actors and underrepresented regions. Continuity mechanisms: lightweight secretariat or coordination hub to track progress, ensure accountability, and maintain momentum between Dialogues. The goal is a Dialogue that is inclusive but execution-focused, linking policy, technical practice, and real-world deployment into measurable progress.

Spanish-speaking and Global South ecosystems (Latin America, parts of Africa): often present as adopters, not co-creators of standards or systems. Practitioners (educators, SMEs, public sector operators, builders): those deploying AI at scale but rarely shaping governance. Open-source communities: key drivers of innovation and access, yet underrepresented in formal policy spaces. Linguistic and cultural minorities: affected by model biases and underrepresentation in datasets. Youth and future workforce: those most impacted by long-term AI transformations. --How to include them-- Funded participation: travel grants, remote-first access, and compensation for contributors from underrepresented regions. Language accessibility: multilingual submissions, interpretation, and support for non-English inputs. Structured practitioner channels: dedicated tracks for real-world deployment case studies and feedback loops into policy. Open calls and public repositories: allow communities to contribute datasets, benchmarks, and proposals transparently. Partnerships with grassroots networks: leverage existing communities (e.g., education and builder networks) to scale participation. Continuous engagement: not only annual events, but ongoing working groups with diverse representation. Inclusion should be designed as infrastructure, not an afterthought.

Interactive, outcome-driven, and continuous engagements, like: Working sprints: short, focused sessions where mixed stakeholder groups produce concrete outputs (e.g., draft standards, benchmarks, policy templates). Policy–practice labs: pair policymakers with practitioners to test how regulation works in real deployment scenarios. Global distributed hackathons: coordinated across regions (including Spanish-speaking ecosystems) to build open tools, datasets, or evaluations aligned with Dialogue priorities. Red-teaming and simulation exercises: stress-test AI systems or governance approaches in real time, involving technical and non-technical participants. Living repositories: open, versioned platforms where contributions (standards, datasets, case studies) evolve continuously between meetings. Regional nodes: local events feeding into the global Dialogue, ensuring context-specific input and broader participation. Lightning use-case sessions: fast-paced presentations from startups, educators, and public sector teams showing real deployments and lessons learned. Deliberative forums: structured citizen or stakeholder assemblies to capture societal perspectives on key issues.

From the experience of Saturdays.ai, a global, community-driven AI education and builder network, and NGO that I founded, these: Hands-on, responsible AI education at scale: Programs that combine technical training with ethics, safety, and real-world deployment help translate governance principles into practice. Participants build AI solutions while integrating considerations such as bias, privacy, and societal impact from the start. "Learning-by-building" as governance infrastructure: Requiring teams to develop end-to-end AI projects (often addressing social or public-sector challenges) creates practical understanding of risks, limitations, and verification needs, bridging the gap between policy and implementation. Distributed, inclusive talent networks: A decentralized model operating across cities and countries enables participation beyond major tech hubs, supporting more equitable access to AI capabilities and aligning with capacity-building goals. Open knowledge and peer learning: Sharing materials, projects, and best practices openly fosters transparency and accelerates diffusion of responsible AI practices, especially in underrepresented ecosystems. Local adaptation with global coordination: Programs are globally connected but locally contextualized (e.g., language, sector needs), which is critical for culturally relevant and effective AI deployment. Multi-stakeholder collaboration: Engagement with startups, public institutions, educators, and industry partners creates feedback loops between builders and policymakers, informing more grounded governance approaches. Low-cost, accessible tooling: Leveraging widely available platforms (e.g., cloud notebooks, open models) lowers barriers to experimentation while promoting reproducibility and auditability. To wrap up, not only about regulation, but also about building capacity, embedding responsibility in practice, and enabling broad participation in AI development.