Towards a Johannesburg declaration on AI governance

The development of artificial intelligence is outpacing global governance efforts. The G20 Johannesburg Summit offers an opportunity to bridge competing visions of AI governance and demonstrate that multilateral efforts can keep pace with the rate of technological change. 

Such efforts to date generally fall into two categories: first, broad-based initiatives at the United Nations aimed at building consensus among its many members and, second, narrower, issue- or context-specific processes at platforms such as those with the G7. Efforts in the first category, such as the UN General Assembly’s decision in August 2025 to establish the Independent International Scientific Panel on AI and the Global Dialogue on AI Governance, reflect the characteristics of ‘universal governance’. They are designed to be inclusive and consensus oriented, with the panel providing independent scientific advice and the dialogue creating a forum for broad multilateral engagement. These initiatives go a step beyond purely declaratory statements, but risk being slow to respond to global developments if not adequately resourced, given that AI development is well beyond its infancy.

Efforts in the second category, such as the code of conduct that emerged from the 2023 G7 Hiroshima Process, offer more concrete pathways for advancing key aspects of AI governance. They nevertheless face familiar criticisms of ‘club governance’ including insufficient inclusivity, the risk of bloc formation and the potential to prompt the creation of parallel organisations. Although not framed explicitly in these terms, China’s recent proposal for a new global AI cooperation organisation focusing on the Global South’s interests can be interpreted as a counterpart to the G7.

Managing the risks, advancing global AI governance

However, such initiatives do not discount the progress made in multilateral efforts at global AI governance. Indeed, multilateral progress in a highly technical and multifaceted domain such as AI governance deserves due recognition. Yet, given the current trajectory, two significant risks confront the international community. The first is the ‘fragmentation risk’, where duelling global AI governance regimes – one centred on the G7 and the other on Chinese-led efforts – could divide global standard setting and disrupt AI value chains. The second is the ‘hollowness risk’, whereby newly created mechanisms such as the Scientific Panel and Global Dialogue risk remain under-resourced and fail to generate the changes needed for global AI governance. Additionally, as private sector acctors remain the primary developers and deployers of AI and their practices often set de facto standards, the risks of both fragmentation and hollowness will be amplified should they not be meaningfully engaged.

This is where the G20 can play an indispensable role. It includes the advanced economies of the G7, the BRICS economies and other influential middle powers, thereby making it more globally representative. As the only platform where competing blocs convene at the same table, the G20 has the potential to serve as a genuine space for convergence – provided its leaders can muster the political will to prevent the emergence of splintered global AI governance regimes. The G20 also possesses a distinct advantage in terms of practical leverage as G20 members are both AI developers and large markets. Furthermore, by incorporating private sector perspectives, the G20 can also ensure its commitments are reinforced by the actors shaping AI’s growth trajectory. If the G20 coordinates effectively on global AI governance, it would not only prevent fragmentation but also provide a foundation for broader multilateral alignment, including within the UN system.

a unified framework for inclusive and accountable AI

Although challenging, the agenda-setting power of the G20 should not be underestimated. It has proven itself capable of shaping ambitious global agendas, such as advancing financial regulation after the 2008 global financial crisis or reinforcing momentum for the 2015 Paris Agreement on climate change. In addition, the G20 does not operate in isolation. A range of technical bodies, such as the International Telecommunication Union and ISO, can help translate high-level declarations into the capacity-building initiatives needed for effective global AI governance.

G20 leaders must therefore work towards a Johannesburg declaration on AI governance by committing to concrete next steps. Such a declaration should not be merely aspirational. It must outline clear priorities for global AI governance such as safety, transparency and accountability. It should also recognise the need to support capacity building in developing contexts and align global AI governance aspirations with the private sector’s objectives. In parallel, leaders should mandate the development of detailed action plans to operationalise the objectives articulated in the UNGA resolution, including how G20 members will engage with and support the Scientific Panel and Global Dialogue. A forward-looking declaration issued at the 2025 Johannesburg Summit will demonstrate how the G20 can deliver on practical convergence and meaningful progress when it comes to global AI governance.