UN-Centered Framework Proposed for Global AI Safety

In a pivotal move toward redefining how the world manages artificial intelligence, a new global framework has emerged calling for a UN-centered governance system to address the accelerating risks of AI. Spearheaded by China and backed by a growing consensus among international experts, the proposal—detailed in the report Advancing a Global Framework for AI Safety and Governance for the Well-being of Humanity—lays out a structured, actionable plan to overcome the current fragmentation in AI oversight. At its core, the framework diagnoses four systemic failures: unpredictable technological evolution, uneven global development, unclear stakeholder responsibilities, and geopolitical competition. In response, it proposes a four-pillar architecture centered on safety, inclusiveness, responsibility, and multilateral coordination. This effort matters because it transforms abstract calls for “global AI governance” into concrete institutional mechanisms, signaling a shift in how nations may soon treat AI—not as a national tech race, but as critical global infrastructure requiring collective stewardship.

The urgency is clear. As AI systems grow more capable, incidents such as cyberattacks enabled by generative models, election-disrupting deepfakes, and algorithmic dependencies in healthcare and energy reveal that AI risks are already transnational. Existing governance remains largely national, reactive, and inconsistent. The report argues that without coordinated action, disparities in regulation could lead to regulatory arbitrage, safety gaps, and escalating geopolitical tensions over frontier AI systems. By reframing AI safety as a global public good—akin to climate stability or pandemic preparedness—the framework positions international cooperation not as idealism, but as operational necessity.

A Layered Approach to Risk and Responsibility

What sets this framework apart is its three-tiered risk model, which disentangles overlapping concerns often blurred in policy debates. The first layer focuses on technical safety: ensuring AI systems are robust, reliable, and free from catastrophic failure modes, particularly in high-stakes domains like nuclear command or aviation. The second addresses societal harm, including bias, misinformation, labor displacement, and erosion of privacy. The third, and most consequential, is systemic risk—the potential for AI to destabilize economies, undermine democratic processes, or concentrate power in unaccountable hands.

This stratification enables targeted interventions. For instance, technical safety might require standardized testing protocols for foundation models, while societal harms call for content traceability and redress mechanisms. Systemic risks demand higher-level coordination, possibly involving restrictions on compute capacity or export controls on advanced chips. The report urges major economies to lead in mutual recognition of safety certifications, akin to aviation or pharmaceutical standards, to reduce duplication and build trust. It also recommends establishing a global risk classification system to enable precise, cross-border identification and response—similar to the International Health Regulations used during pandemics.

Institutionalizing Science and Dialogue at the UN

To operationalize these ideas, China’s Global AI Governance Action Plan, presented to the United Nations, proposes two new mechanisms under the UN umbrella: the Independent International Scientific Panel on AI and the Global Dialogue on AI Governance. The scientific panel would function like the Intergovernmental Panel on Climate Change (IPCC), synthesizing peer-reviewed research to inform policy with authoritative assessments. It would evaluate emerging risks, monitor compliance with safety standards, and issue early warnings—free from political interference.

The Global Dialogue, meanwhile, would serve as a multistakeholder forum bringing together governments, tech companies, civil society, and academia. Unlike closed-door negotiations, this platform aims to foster transparency and inclusivity, especially for low- and middle-income countries often excluded from AI governance discussions. Both mechanisms are designed to complement—not replace—national regulators, respecting sovereignty while enabling coordination. As stated in the Action Plan, “Pooling scientific knowledge is most efficient at the global level, enabling joint investment in a global public good, and public interest collaboration across otherwise fragmented and duplicative efforts.”

Bridging the Global AI Divide

A central concern woven throughout the framework is equity. The report warns of a widening “AI divide,” where a handful of countries and corporations control the most powerful systems, leaving others dependent and vulnerable. To counter this, the proposal includes a call for a global fund for AI development, modeled on initiatives like the Green Climate Fund, to support capacity building in underserved regions. This would finance local AI research, workforce training, and infrastructure deployment, ensuring that benefits are broadly shared.

Additionally, the framework advocates for an international network of AI standards exchanges, allowing countries to harmonize regulations and share best practices. It emphasizes that inclusiveness is not merely ethical—it enhances safety. Systems trained on diverse data and governed through pluralistic input are less prone to blind spots and more resilient to misuse. As one supporting analysis notes, treating AI governance capabilities as “shared resources” strengthens collective security, much like open scientific collaboration during health emergencies.

Geopolitical Realities and Pathways Forward

Despite its ambition, the proposal faces significant hurdles. Geopolitical fragmentation remains a core challenge. While China champions a UN-led model emphasizing sovereignty and non-interference, Western powers have pursued alternative paths, such as the U.S.-led Declaration on Responsible AI or the EU’s binding AI Act. These divergent approaches reflect deeper disagreements over data rights, surveillance norms, and the balance between innovation and control.

Yet there are signs of convergence. The emphasis on risk classification, safety testing, and emergency response appears in multiple frameworks, suggesting common ground. The call for mutual recognition of certifications could become a pragmatic starting point, reducing friction in trade and deployment while building confidence. Moreover, the neutrality of the UN platform offers a rare space for dialogue amid rising tech tensions.

Experts note that success will depend on whether the proposed scientific panel gains genuine independence and technical credibility. Past UN bodies have sometimes been hampered by politicization or underfunding. To avoid this, the framework suggests early operationalization with seed funding from willing states and philanthropies, alongside transparent governance rules.

As AI continues to evolve faster than institutions can adapt, the push for a coordinated global response has never been more urgent. This framework does not offer a final solution, but it provides a roadmap—one that moves beyond rhetoric to institutional design. By anchoring governance in science, equity, and multilateralism, it reframes AI not as a zero-sum contest, but as a shared challenge demanding collective action. The coming years will test whether nations can rise to that task. If they do, this moment may be remembered as the beginning of a new era in global technology governance—one where humanity governs AI, rather than the other way around.