Ivannikov Institute for System Programming of the Russian Academy of Sciences (ISP RAS)

The success of the first session should be measured not by declarations, but by the creation of a sustainable mechanism for practical cooperation among governments, science, business, and civil society. Key outcomes could include: 1. Agreement on a common conceptual framework — basic terms, principles, and AI risk classifications necessary for the comparability of national approaches. 2. Adoption of a roadmap for international cooperation for 2026–2028 with specific priorities: AI safety, computing infrastructure, standards, workforce development, and trusted data. 3. Launch of a scientific and expert exchange network among research centers and international organizations for the independent assessment of technological risks and opportunities. 4. Launch of a dedicated international "AI for Science" track focused on the use of AI in fundamental research, medicine, climate, new materials, energy, agriculture, and other fields. Such a track could bring together research centers, universities, technology companies, and international organizations around practical challenges. 5. Support for countries with emerging AI ecosystems through programs for knowledge transfer, specialist training, and access to computing resources. 6. Development of regulatory interoperability mechanisms to reduce fragmentation of requirements and barriers to international research and digital trade. 7. Recognition of multilingualism and cultural diversity as essential elements of the global AI order. 8. Establishment of a regular progress monitoring procedure, including an annual report on the state of global AI governance. The ultimate result should be a shared understanding that global AI governance is not about restricting innovation, but about making it safe, fair, and accessible for all countries.

• Safe, secure and trustworthy AI
• AI capacity-building
• Transparency, accountability, and human oversight
• Open-source software, open data and open AI models

For a research organization, the priority is to ensure that AI development is accompanied by engineering reliability, scientific reproducibility, and public trust. Safe and trustworthy AI is a fundamental issue. Without methods for verification, testing, resilience against attacks, and quality control, large-scale deployment of AI in industry, public administration, healthcare, and education is not possible. Capacity-building is critical for narrowing the global technological gap. It requires skilled professionals, research infrastructure, computing capacity, and modern educational programs. Transparency, accountability, and human oversight are essential for the use of AI in sensitive sectors. Algorithmic decisions must be verifiable, and responsibility must be clearly defined both legally and organizationally. Open-source software, open data, and open models accelerate scientific progress, lower barriers to entry for universities and small companies, enable independent auditing, and foster a competitive innovation environment. The combination of these four areas creates a sustainable AI ecosystem: safe, accessible, and oriented toward public benefit.

Yes, a number of issues require separate attention. 1. Computational sovereignty and access to infrastructure. AI development is increasingly dependent on access to computing resources, specialized processors, and cloud platforms. This is becoming a factor of global inequality. 2. Energy and environmental sustainability of AI. The growth of large models is increasing the energy consumption of data centers. Metrics and standards for energy efficiency are needed. 3. Scientific reproducibility and independent auditing of models. Many modern systems are opaque and difficult for the academic community to reproduce. 4. Linguistic and cultural diversity. Most resources are concentrated around a limited number of languages. Many of the world's languages lack high-quality data and models. 5. AI for science and public good. The use of AI in fundamental science, climate, healthcare, emergency response, and education should be considered as a separate area. 6. AI for science and public good. The application of AI in both fundamental and applied science should be addressed separately: generating scientific hypotheses, discovering new materials and medicines, climate modeling, biomedicine, agricultural technologies, engineering design, and analysis of large-scale scientific data. This area is important not only as a technology market, but also as a mechanism for accelerating solutions to global challenges such as health, food security, energy, climate, and technological sovereignty. 7. Market concentration and competition. The high cost of computing strengthens the dominance of a limited number of companies, affecting access to technologies. 8. Long-term labor market sustainability. It is important to focus not only on job displacement, but also on creating new qualification and career pathways.

For Russia and the Eurasian region, the key issues are technological independence, workforce development, and access to advanced AI solutions. Main challenges: 1. A shortage of computing resources and hardware infrastructure, which slows the development of competitive models. 2. Rapidly growing demand for qualified specialists in AI engineering, microelectronics, security, and data governance. 3. Fragmentation of standards and regulation among countries, which complicates cross-border projects. 4. Insufficient representation of the Russian language and other regional languages in global models and datasets. 5. Cyber risks and misuse of AI, including fraud and the generation of harmful content. 6. Insufficient readiness of scientific data for AI use: fragmented archives, lack of unified metadata standards, and difficult access to experimental data. Opportunities: 1. Leveraging strong mathematical traditions and engineering education to develop trustworthy AI. 2. Applying AI in industry, transport, energy, healthcare, and the public sector. 3. Creating regional open language models for the multilingual Eurasian space. 4. Developing international scientific consortia and joint data platforms. 5. Exporting solutions in industrial AI and cybersecurity. 6. Using AI to accelerate research in priority fields: new materials, biomedicine, energy, agriculture, climate, space, and engineering design. This can shorten the cycle from scientific hypothesis to technological solution and increase returns on R&D investment. With effective governance, AI can become a driver of productivity growth and economic modernization.

The Global Dialogue can become a central neutral platform where trust is built among governments and key participants in the AI ecosystem. It can serve the following functions: 1. Reducing regulatory fragmentation through the exchange of approaches and the development of compatible principles. 2. Scientific diplomacy — connecting research centers from different countries to work on AI safety, interpretability, and testing. 3. Supporting developing countries through programs for skills development, infrastructure, and access to data. 4. Early identification of global risks — discussing emerging threats related to AI deployment and methods for their prevention. 5. Promoting standards and best practices agreed upon across different regions. 6. Shaping a common agenda for AI for public good: healthcare, climate, education, and sustainable development. The Dialogue's particular value lies in its inclusiveness and its ability to bring together different regulatory models without imposing a single system.

The Dialogue should build on existing international mechanisms: 1. Initiatives within the United Nations system (International Telecommunication Union, UNESCO, UNDP); 2. Standardization bodies such as ISO/IEC, IEEE, and ITU-T; 3. Regional regulatory practices of BRICS, G20, APEC, and others; 4. Academic networks and scientific panels on AI safety; 5. Industry partnerships involving major technology companies and research centers. Added value of the Global Dialogue: • Bringing fragmented processes together into a single global space for discussion; • Creating equal opportunities for all countries to participate in shaping international approaches to AI; • Balancing the interests of governments, business, and academia; • Linking rulemaking with the real state of technology; • Regular monitoring of the implementation of recommendations. The Dialogue should not duplicate existing institutions, but rather serve as a "coordination hub".

Stakeholder contributions: • Governments – regulatory approaches, national strategies, and implementation mechanisms. • Science and universities – independent expertise, risk assessment methodologies, and workforce training. • Business – technological practice, investment, and scalable solutions. • Civil society – human rights, inclusiveness, and public oversight. • Open-source software developers – transparency and accessibility of tools. • International organizations – coordination and comparability of practices. Recommended format: 1. High-level plenary sessions. 2. Thematic tracks with concrete deliverables. 3. Scientific and technical expert forum. 4. Regional roundtables. 5. Youth and university track. 6. Permanent digital platform between sessions. 7. Public registry of proposals and outcomes. 8. Creation of a permanent digital platform for discussions and information exchange. 9. A dedicated scientific and technological "AI for Science" track, including presentations of research case studies, discussions on scientific data standards, reproducibility, computing infrastructure, and international joint research programs. It is important that each track produce not only discussions, but also concrete recommendations, draft standards, and cooperation programs.

Underrepresented groups: 1. Low- and middle-income countries; 2. Researchers from universities outside global centers; 3. Speakers of small and underrepresented languages; 4. Industry engineers and practitioners; 5. Educators and the medical community; 6. Youth; 7. Small and medium-sized enterprises. How to include them: • Grant support for participation and travel; • Full remote participation options; • Work in the six official United Nations languages and expanded language support; • Open calls for proposals; • Regional preliminary consultations; • Participation quotas for underrepresented groups on panels; • Special youth and scientific sessions. Without broadening participation, global AI governance risks reflecting the interests of only a limited number of actors.

1. Policy labs – joint real-time drafting of recommendations. 2. Red teaming sessions – testing AI risks through adversarial exercises. 3. International public-good hackathons focused on healthcare, climate, and education. 4. Marketplace of solutions – showcasing ready-made tools for auditing, testing, and training. 5. Peer review sessions of national AI strategies. 6. Data collaboratives – mechanisms for the secure sharing of data. 7. Permanent virtual working groups operating between annual meetings. 8. Creation of a repository of trusted solutions and trusted software/AI. It is important to establish an open, professionally moderated repository of trusted system and AI components, including verified versions of core software and ML frameworks. Such an approach reduces supply-chain compromise risks, improves reproducibility, and accelerates the adoption of safe technologies in industry and the public sector. 9. AI for Science Grand Challenges – international research competitions and pilot programs focused on high-impact public-interest challenges such as new medicines, materials, climate risks, food security, and energy. 10. Federated science data rooms – secure environments for collaborative work with scientific data. These formats transform a conference from a discussion platform into a mechanism for joint problem-solving.

1. Risk-based regulation. Stricter requirements for high-risk AI applications and simplified rules for low-risk solutions. 2. Mandatory AI impact assessments before deployment: safety, citizens' rights, data quality, resilience, and the possibility of human oversight. 3. Regulatory sandboxes that allow innovative solutions to be tested under the supervision of governments and experts. 4. Independent auditing of models and algorithms, including checks for vulnerabilities, discriminatory effects, and resilience to manipulation. 5. National workforce development programs - from school-level education to retraining specialists for the economy and public administration. 6. Support for open-source ecosystems and open scientific repositories of data and models, while maintaining security requirements. 7. Russia's National AI Development Strategy provides an important benchmark: combining technological leadership, trust in AI, talent development, and the deployment of AI across the economy and social sectors. It is an example of a strategic approach where innovation is combined with security and sovereign infrastructure considerations. 8. International standardization of metrics for reliability, interpretability, and energy efficiency of models. Effective AI governance should rely not only on rules, but also on a technological infrastructure of trust: standards, expertise, repositories of verified solutions, and systems of accountability.