Research Center for Artificial Intelligence at MEPhI

The success of the first meeting of the UN Global Dialogue on AI Governance will be determined by the transition from declarations to the creation of practical coordination mechanisms. Key indicators of success will be: 1. Institutional Clarity Develop a clear roadmap for the creation of an International Scientific Panel on Artificial Intelligence (IPCC for climate change). Success depends on agreeing on its mandate, principles of independence, and mechanisms for annual reporting to participating countries. 2. Inclusiveness and bridging the digital divide Concrete commitments to creating funds or capacity-sharing networks for the Global South. If the dialogue is limited to the interests of "tech giants," it will fail. A mechanism that allows developing countries to participate in governance, rather than simply consume ready-made solutions, will be successful. 3. Ethical and Regulatory Compatibility Adoption of a framework document that does not replace national laws but establishes "red lines" for example, in the area of social credit. Consensus in defining risks is important: from catastrophic to everyday ones (discrimination, disinformation). 4. Interoperability Standards Agreement to develop unified technical standards for content security and labeling. This will allow different regulatory models (such as those of the EU, US, or China) to communicate with each other, preventing fragmentation of the global market. The essence of success: If the meeting results in the establishment of a permanent secretariat and the approval of a schedule of specific working sessions on AI safety, the Dialogue can be considered successful. The main goal is to transform the UN into a central hub where technical expertise meets political will.

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

For the MEPhI Research Center for Artificial Intelligence, as an organization specializing in advanced developments in AI for transport and critical infrastructure (e.g., V2X systems and predictive analytics), the following four blocks are the highest priority: 1. Safe, Secure, and Trusted AI This block is the foundation for implementing AI in the real economy. When it comes to driverless transport or managing urban infrastructure, the cost of error is critically high. The organization's involvement here is justified by the need to develop standards for fault-tolerance and protection of algorithms from external interference, which directly impacts technological sovereignty and the physical safety of citizens. 2. Building AI Potential This priority relates to educational and research missions. Active participation in this area allows us to scale up expertise, train personnel for intelligent transport systems, and bridge the technological gap. This strategic effort integrates scientific advances into real-world logistics and production chains. 3. AI Transparency and Accountability, Human Oversight For trusted AI systems, it is essential that algorithmic decisions be interpretable. In complex engineering systems, a "black box" is unacceptable. Focusing on human oversight ensures that AI remains a decision-support tool, retaining final control in emergency situations. 4. Open-source software, open data, and open AI models Working with open standards and data (e.g., AIXM 5.1 in aviation or V2X protocols) is critical to ensuring interoperability. Support for open models accelerates innovation and enables code security verification by a community of experts, which is especially important when creating complex cross-platform solutions for smart cities. These priorities balance the stringent security requirements of critical systems with the need for open international cooperation and human resource development.

Despite the comprehensive nature of the list in Resolution 79/325, the rapid development of technologies in 2024-2026 highlights three critical aspects that remain in a "grey area" or require separate categories: 1. Environmental Sustainability and Green AI Energy consumption and carbon footprint are not on the list. Training giant models and running data centers require colossal resources. Global dialogue must address the environmental impact of AI to ensure that technological developments do not conflict with UN climate goals. 2. Legal Liability and Intellectual Property Status While "technical consequences" have been mentioned, the issue of legal authorship of AI-generated content and liability for damage caused by autonomous systems (for example, in logistics) requires further analysis. Global rules for distributing liability between the developer, owner, and user of the algorithm are needed. 3. Biodigital Convergence and Neurotechnology The development of brain-computer interfaces and the use of AI in synthetic biology create risks that go beyond traditional digital ethics. Issues of cognitive freedom and the protection of neurodata (protection from unauthorized access to thoughts or biological processes) are becoming urgent. 4. Geopolitical Resilience of Critical Infrastructure In the context of internet fragmentation, the question of "digital data sovereignty" arises. How can global transportation systems (V2X, aviation) be ensured if AI security protocols are nationalized? This issue of tech diplomacy is critical for the smooth functioning of global logistics. Taking these topics into account will transform the Global Dialogue from a platform for discussing software into a mechanism for regulating the material world, where AI directly impacts the environment, property rights, and human biological safety.

Gaps in AI governance create a "legal vacuum" effect, which will be particularly acute in the transport and urban infrastructure sectors in 2026. Challenges • Technological fragmentation: The lack of unified standards for "safe and trusted AI" hinders the cross-border movement of autonomous vehicles. Differences in algorithm certification between the EAEU and BRICS countries create barriers to the export of high-tech solutions (e.g., V2X systems). • Trust deficit: Without transparency and accountability (human oversight), the implementation of AI in critical systems faces social resistance and legal uncertainty when investigating road accidents or man-made incidents. • Skill drain: Gaps in "capacity building" lead to the concentration of intellectual property among global vendors, putting national research and educational centers in a position of playing catch-up. Opportunities • Advanced regulation: Addressing gaps through the implementation of "open models and data" enables the creation of local ecosystems independent of Western proprietary platforms. This is an opportunity for the region to become a leader in Trusted AI for industrial applications. • Economic impact: Clear governance rules allow businesses to invest in predictive maintenance and smart logistics, reducing operating costs by 15-20% by minimizing downtime and accidents. • Standards export: The first country to implement reliable AI model verification mechanisms can scale its protocols to the regional market (Central Asia, Eastern Europe), creating a technological security belt. Bottom line: Success in the selected blocks will transform AI from an "experimental add-on" into a reliable foundation for the national economy. The key opportunity here is the transition from purchasing foreign licenses to creating our own sovereign security standards.

The UN Global Dialogue acts as a universal moderator, capable of transforming disparate national strategies into a coherent international architecture. Its role in fostering cooperation is manifested in three aspects: 1. Creating a unified "conceptual framework" Dialogue removes semantic barriers. For international cooperation, it is critical that terms like "trusted AI" and "high risk" are interpreted consistently across jurisdictions. The UN creates a common ground on which countries with polarized political systems can reach agreements. 2. A Platform for Inclusive Knowledge Transfer Unlike closed clubs (such as the G7 or G20), the Global Dialogue ensures the participation of developing countries. It serves as an intellectual hub where industry leaders share expertise in exchange for access to new markets and data. This transforms AI governance from a tool for competition into a tool for collective development. 3. Synchronization of regulatory sandboxes The dialogue facilitates the creation of mechanisms for the mutual recognition of security standards. This helps avoid legal fragmentation, where companies must adapt their models to the requirements of each individual country. Cooperation here aims to create a seamless digital space for innovation. 4. Monitoring global existential risks The UN takes on the role of "early warning." Global dialogue allows countries to jointly invest in AI safety research that is too expensive or complex for a single state. Key role: Transforming the AI "technology race" into a "standards and security race." Dialogue legitimizes global rules of the game, ensuring that technological progress does not undermine the stability of the international system.

The UN's global dialogue should not reinvent the wheel; its goal is to become a "system of systems" that brings together disparate efforts. Key Initiatives for Collaboration • International AI Agency: Building on technical security standards developed by leading institutions (e.g., the AI security institutes of the US, UK, and China). • Regional Alliances: Collaborating with the EAEU (on transport corridors and technical regulations) and BRICS (on sovereign data). This will ensure that the interests of different technology groups are taken into account. • Relevant UN Agencies: Building on the work of ICAO (aviation) and ITU (telecommunications) to implement AI in critical infrastructure, such as V2X systems and urban traffic management. The Added Value of the Global Dialogue The Dialogue's primary value is its universal legitimacy, which the G7 or corporate consortia lack. 1. Political Inclusiveness: Only under the auspices of the UN do developing countries gain a real voice in governance, preventing AI from becoming a tool of a new "digital colonialism." 2. Arbitration and Interoperability: The Global Dialogue can create a mechanism for the mutual recognition of security certificates. If an AI system is recognized as "trusted" at the UN level, this simplifies its export and use in international supply chains. Bottom line: The added value lies in transforming the fragmented landscape of initiatives into a unified ecosystem where technological standards are inextricably linked to international law and human rights.

The effectiveness of the Global Dialogue depends directly on whether it becomes a platform for real interaction, and not simply a platform for officials. Stakeholder Contributions • Academia and research institutes: Should provide "scientific grounding" by acting as independent auditors of technological risks and developers of methodologies for "trusted AI." • Private sector and industry leaders: Their role is to provide technical data, participate in the creation of sandboxes for testing cross-border systems (e.g., V2X), and invest in infrastructure in the Global South. • Civil society: Acts as a guarantor of respect for human rights, ethical standards, and linguistic diversity, preventing algorithmic bias. Recommendations for format and structure 1. Hybrid "Core-Industries" structure: Establish a permanent secretariat (core) and mobile industry working groups (transport, medicine, security). This will avoid generalities and focus on specific technical protocols. 2. "Dynamic Sandbox" format: Instead of static reports, hold regular sessions to share AI implementation cases. Use interactive dashboards to monitor global progress in real time. 3. Two-tier decision-making: o Expert level: Consensus of engineers and ethicists ("physicists" and "lyricists") on technical standards. o Political level: Approval of these standards by UN member states. 4. Rotational principle: Meetings should be held not only in New York or Geneva, but also in technology hubs in different regions (Singapore, Nairobi, Astana) to ensure inclusiveness. Main format: Transition from "monologue conferences" to "management hackathons," where the outcome of each meeting is a specific draft standard or memorandum on technical compatibility.

Current discussions are clearly biased toward the "technological optimism" of developed countries and the interests of large corporations. For the Global Dialogue to be truly inclusive, the following groups must be engaged: 1. Representatives from the Global South and Developing Economies Standards are currently being written where computing power is concentrated. The voice of countries in Africa, Latin America, and Central Asia is essential to avoid "digital colonialism." • How to include: Quotas for seats in working groups and the creation of funds to cover participation fees for experts from these regions. 2. Small and medium-sized businesses (SMBs) and open-source developers Strict regulatory requirements are often only met by giants (Google, OpenAI). The opinions of independent developers and startups are ignored, stifling competition. • How to enable: Conduct specialized industry tracks and consultations to assess the impact of regulation on small businesses. 3. Regional Research and Education Centers The voice of the academic community is often drowned out by lobbyists. Researchers from technical universities (for example, those working on applied AI in engineering or transportation) have unique expertise in local safety. • How to enable: Create a UN Network of Competence Centers, where universities can directly submit technical amendments to standards. 4. Linguistic and Cultural Communities AI is trained primarily on English-language data, which leads to the degradation of minority languages and cultural codes. • How to enable: Integrate linguistics and anthropology specialists into ethics groups to develop requirements for multilingual models. Inclusion mechanism: Instead of ad hoc forums, a digital platform for collective participation should be implemented, allowing any accredited organization to submit proposals in real time. Success is only possible if AI governance becomes a "Wikipedia of meanings" rather than a closed club of select countries.

To prevent the Global Dialogue from becoming a series of static reports, it is necessary to introduce formats that stimulate practical interaction and rapid iteration of ideas. 1. Thematic "regulatory hackathons" Instead of discussing theoretical risks, participants are asked to develop a prototype technical standard or draft memorandum for a specific industry (for example, a data exchange protocol for V2X) within 48 hours. This shifts the discussion from a political to an engineering and legal perspective. 2. "Regulatory Sandboxes in Real Time" Format Creating digital simulations where delegates can see how their proposed governance regulations will impact the economy and innovation. • Example: Visualizing how mandatory model certification will slow down startup development in low-GDP regions. 3. Reverse Pitching Governments don't report on their achievements, but instead "sell" their unsolved AI problems (for example, a lack of personnel or biased algorithms in government services). Tech giants and research institutes, in turn, offer ready-made architectural solutions or resources to address them. 4. Red Teaming for Politics Special groups of experts (ethicists, hackers, futurologists) publicly "attack" proposed draft resolutions, identifying loopholes for monopolies or risks to human rights. This ensures stress testing of decisions before they are officially adopted. 5. Digital "doubles" of the Global Dialogue A platform for continuously collecting proposals from the global community during the intersessional period. AI analyzes thousands of proposals from scientists and small and medium-sized businesses, creating a focused agenda for in-person meetings. Innovation in structure: The key value will be the transition to a modular approach, where the outcome of the meeting is not a single cumbersome document, but a package of "plugins"—specific technical and ethical protocols ready for immediate implementation into national legislation.

Effective AI governance today relies on a combination of strict regulation, industry standards, and automated monitoring platforms. Here are the most significant examples for 2026: 1. Regulatory Anchors and Standards • National Standards for Trusted AI (Russia): State Standards for Neural Network Testing (GOST R 70462.1-2022/ISO/IEC TR 24029-1-2021), which emphasize the interpretability of decisions, which is critical for predictive analytics and V2X systems. 2. Practical Approaches • Human-in-the-loop: A principle enshrined in urban traffic management, where AI optimizes traffic light cycles, but critical infrastructure changes require operator approval. • Regulatory Sandboxes: The practice of temporarily lifting some restrictions to test innovations (such as driverless taxis) under strict government oversight, allowing for safe testing of technologies before their mass launch. Key trend for 2026: Shifting focus from purely ethical discussions to the creation of digital AI passports that contain the entire training and audit history of a model, similar to a car's registration document.