Moscow state university

Launch an Open AI Infrastructure Commons: a voluntary UN-backed system that gives researchers from developing countries fair access to computing power and curated data, with transparent, non discriminatory rules.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Open-source software, open data and open AI models

(1) is the technical bedrock. Without verifiable safety, security, and trustworthiness, any AI deployment risks systemic harm - especially as systems approach autonomy. Building capacity or making systems work together cannot fix an unsafe model. (2) ensures that governance remains grounded in real world diversity. AI does not operate in a vacuum; it reshapes labour, culture, language, and social structures. Ignoring these implications means designing rules that fit narrow contexts while failing everywhere else. (3) provides the universal normative anchor. Concepts like "safety" or "trust" are easily captured by powerful interests. Human rights law (dignity, non discrimination, accountability) gives universal, legally binding protection that works across cultures and outlasts tech trends. (4) is the structural condition for inclusive innovation. Open source software, open data, and open models prevent monopolisation of AI capabilities. Without openness, even well intentioned governance becomes a tool for incumbent concentration. Openness is not a luxury; it is a prerequisite for distributed oversight and democratic participation. Together, these four clusters form a coherent action framework: safe systems (1) grounded in real impact analysis (2), anchored in human rights (3), and built on open foundations (4).

-

In the research sector, the absence of global verification protocols creates fragmentation of knowledge: laboratories duplicate risk assessments without trusting each other's results, wasting resources. The key challenge is asymmetric access to closed models. Industry and developed countries analyse frontier systems internally, while academia in developing countries works with black‑box models. This creates inequality not only in using technology, but also in understanding it. The opportunity is a global network of trusted computing nodes under UN mandate, where researchers from any country obtain verifiable, privacy‑preserving access to reference models. Harmonised transparency requirements for training data (not only for big tech) would enable reproducibility and reveal hidden biases. Governance built on trust, not barriers, can become the greatest catalyst for science in AI.

The Global Dialogue must act as the core framework (like an operating system) for AI governance. Three unique roles: 1. Today, regulators in the EU, US, China, India, Russia…speak different languages. The Dialogue can create a technical committee establishing equivalence requirements - a «Basel Committee for AI». (This proposal draws direct analogy to the Basel Committee on Banking Supervision (BCBS)) 2. Early warning mechanism for force majeure. If a country faces an uncontrollable AI incident (e.g., data poisoning attack or agent escape), the Dialogue provides cryptographically secure alert channels, joint investigation protocols, and coordinated rollback – a «nuclear hotline for algorithms». 3. Create a «norm factory» that produces living, version controlled guidelines (e.g., Geneva Specification for Agentic Systems), updated every two years by experts - instead of fixed treaties that AI outgrows quickly. Without this functional architecture, the Dialogue becomes only another talk shop.

Its unique value is universal legitimacy, backed by the UN, plus a clear mandate for interoperability. No other platform can bring together a chip designer from China, ML engineers from USA, and a digital minister from Russia. Most critically, the Dialogue can create an Advisory Body on Long‑term Existential Risks (as an adjunct to the Scientific Panel), which no other process currently does. This addresses human survival in the age of superintelligence, not just compliance.

Each group can bring essential assets: - Industry- deployment experience and resources for sandboxes. - Civil society - rights advocacy and voice for vulnerable groups. - Academia - evaluation methodologies and long‑term modelling. - Technical community - verifiable protocols.

Two voices are systematically excluded: - Persons with disabilities and neurodivergent individuals. AI already shapes their environment, but their lived experience of assistive technologies and discrimination risks is irreplaceable. Solution: mandatory speaking quotas and funded assistive participation (sign language, alternative communication). - Youth and future generations (those aged 0-18 today, who will live longest with our decisions). Solution: an «Ombudsperson for Future Generations» with the right to flag irreversible risks beyond 2100. Also introduce democratic lotteries to randomly include young people aged 16-25 from non‑represented countries.