On 26 August, the UN General Assembly (UNGA) adopted a new resolution on global AI governance. Resolution A/79/325 sets up two new governance mechanisms: an Independent International Scientific Panel on AI and a Global Dialogue on AI Governance, the latter to be launched during next week’s UNGA.
The UN’s new AI governance architecture is a symbolic triumph. It is also ambitious, by far the world’s most globally inclusive approach to governing AI.
But in practice, the new mechanisms look like they will be mostly powerless. They may end up being hampered by their design, funding uncertainties, an intensifying global AI race, the widening divide between AI ‘haves’ and ‘have-nots’, and disruptive technological leaps.
What will it look like?
The new governance architecture has not come out of the blue. Parts of AI governance have already been internationalized by states, companies and organizations.
Sometimes, this internationalization is manageable. There are some shared priorities in AI governance (like the importance of technical standards for safety and interoperability) and existing institutions to discuss them in. There are also ambitious efforts to fill gaps and build shared understandings.
But some differences – in politics, values and capabilities – are difficult to bridge. AI is notoriously tricky to govern nationally, let alone internationally. And many parts of AI are already too politicized for multilateral progress, such as its military applications (notwithstanding dialogue on its responsible use). The UN’s AI efforts have also been born into an intense global rebalancing: widening capability gaps, the US multilateral retreat and China’s expanding global influence.
The new mechanisms do not aim to enforce a universal approach. Rather, they aim to facilitate coordination and consensus on certain issues while avoiding others. Notably, the use of AI in the military domain is explicitly out of scope, which raises questions about how dual-use technology will be treated.
The new scientific panel will be formed of 40 global experts and led by two co-chairs (one from a developed country, the other from a developing country). It is tasked with producing an annual scientific report – a global snapshot – on the risks, opportunities and impact of AI.
The new global dialogue will convene states and multi-stakeholders – like tech companies and civil society – once a year to discuss issues like AI capacity gaps between countries, transparency and open-source AI models.
The architecture clearly draws lessons from other UN processes, ranging from cybersecurity to climate science – critics and supporters alike have been quick to dub the panel an ‘IPCC for AI’. Including multi-stakeholders in the global dialogue is a good sign, as is its commitment to promoting human rights and compliance with international law.
Importantly, the resolution mandates geographic and gender balance, which – if implemented mindfully – should enable more but not wholly equitable representation. However, the next meetings will be held in Geneva and New York instead of Global South alternatives, which could create financial, travel and visa barriers for some participants.
Three key challenges
The panel and global dialogue are deliberately cautious and designed to withstand AI race dynamics. But bringing parts of AI regulation under UN auspices by no means automatically enables representative, inclusive and effective governance.
As with much UN-based governance these new mechanisms face three key challenges: lack of funding, the turbulence of the US-China rivalry, and enforceability (or lack thereof).
First, the UN’s funding troubles are a major hurdle. While a global fund for AI that would help developing countries access and develop the technology is compelling on paper, there are substantial barriers to its launch – not least amid US cuts to UN funding.
Another cause for concern is that the secretariat for the new mechanisms – the Office for Digital and Emerging Technologies – will apparently depend on voluntary, direct funding, including from tech firms. Although the resolution commits to transparency and avoiding conflicts-of-interest, this has raised concerns among some experts about the mechanisms’ sustainability and ability to maintain independence from political and corporate influence.
Second, as the US pulls back from multilateral organizations, there are concerns about China’s growing influence at the UN. For years, China has positioned itself as the global champion of the Global South when it comes to technology priorities. China worked with the G77 countries in negotiations on the UN Global Digital Compact, promoting principles like state-centric governance, strict sovereignty and non-interference. In 2024, the UNGA passed a China-sponsored resolution on AI capacity-building.
The world will be watching how China navigates this year’s UNGA. It follows China’s recent World AI Conference – where Premier Li Qiang touted Chinese AI solutions and global leadership, with an ambitious action plan rivalling that of the US – and President Xi Jinping’s assembly of autocratic leaders for the SCO summit and Beijing military parade.
Third, while the new AI mechanisms will help set the global agenda on AI risks and opportunities, they have rightly been criticized for their inability to properly enforce it. This is a challenge also faced by other UN initiatives and processes, such as on cybersecurity and internet governance.
Regulating a fast-moving issue like AI under UN auspices presents another difficulty: the institutions might not be able to keep pace with technological capabilities and their disruptions.
Traditional international institutions, like the new AI governance architecture, are increasingly ill-suited to respond to an unpredictable global AI crisis in nuclear or biosecurity. Could they raise the alarm about a looming AI catastrophe – and broker the cooperation needed to prevent or solve it?
Realistic expectations
The new mechanisms will be assessed in 2028. Despite their limitations and the many challenges they will come up against, they should not be overlooked or discarded. Instead, the focus should be on using the mechanisms for their actual – not aspirational – offerings.
