South Africa has gone to this year’s United Nations General Assembly asking for something few countries have managed to agree on: a commitment to build a global framework for governing artificial intelligence. The push comes less than a fortnight after President Cyril Ramaphosa made the same case to BRICS leaders, and while South Africa’s own national AI policy is still being rewritten after its first draft was withdrawn.
Mxolisi Nkosi, South Africa’s Permanent Representative to the UN, said the country wants the 81st General Assembly to produce a statement committing the international community to the development of a framework for the global governance of artificial intelligence, iAfrica.com reported. He warned of how AI can have a devastating impact without guardrails. The Assembly’s high-level week opened in New York on 22 September.
What South Africa is actually proposing
The detail comes from Ramaphosa’s address to the BRICS summit in India on 13 September. He called for an international mechanism to evaluate AI systems independently, for meaningful human control to be preserved, for serious incidents to be reported as a matter of obligation, and for safeguards that get stronger as AI capabilities advance, Ecofin Agency reported. He also asked for external audits and government oversight of the companies building the most advanced systems.
“We should therefore neither fear AI nor approach it recklessly,” Ramaphosa said. “Our responsibility is to ensure that human governance advances as rapidly as artificial intelligence itself.”
The part aimed most squarely at countries like South Africa is the call for investment in sovereign AI capacity in developing economies. The argument running through South Africa’s position is that countries cannot be expected to accept governance rules for technologies they are unable to build or assess for themselves. Pushing the same message through BRICS and the UN at once is a deliberate attempt to give that argument weight in both forums.
The machinery already exists, in part
South Africa is not starting from nothing. The General Assembly established an Independent International Scientific Panel on Artificial Intelligence in August 2025, and its 40 members were appointed in February this year to produce independent assessments of AI’s capabilities, risks and impacts. South African computer scientist Vukosi Marivate, a co-founder of Lelapa AI, sits on the panel. What South Africa wants now is a political commitment that turns assessment into an agreed framework that governments are expected to follow.
The awkward part: the domestic rulebook
The case South Africa is making abroad is stronger than its record at home. The Department of Communications and Digital Technologies published a Draft National AI Policy for public comment on 10 April, proposing a risk-based approach with stricter rules for high-risk uses in healthcare, financial services and law enforcement, law firm Bowmans noted. The draft was later withdrawn after its reference list was found to cite research that does not exist, and a revised version is not expected at Cabinet before November, as we reported earlier this month.
Financial regulators are also waiting. South Africa’s financial sector regulator has said it will hold off on AI-specific rules until a global standard-setting body reports back, as we covered last week. In other words, South Africa is asking the world to move faster on AI governance while, domestically, its own rules are waiting partly on the world.
Why it matters to South African business
For most companies, AI governance arrives through the tools they buy rather than the laws they read. The large models South African firms use are built abroad, and the obligations placed on those developers, such as incident reporting, audits and disclosure, shape what local buyers are told about the systems they deploy. A global framework that includes developing economies in setting those rules would give South African firms more say over the terms on which they adopt AI.
The domestic delay cuts the other way. Without a national policy, banks, insurers, retailers and healthcare providers using AI to make decisions about customers have no South African rulebook beyond existing law, including the Protection of Personal Information Act. Companies deploying AI in high-stakes decisions would be wise to document how those systems work and who is accountable for them now, rather than wait for the policy to force the question.
What to watch
Three dates matter. The first is whether the General Assembly’s outcome documents include the commitment South Africa is asking for. The second is the revised national AI policy reaching Cabinet, expected by November. The third is the first substantive output from the UN’s scientific panel, which will show whether independent international evaluation of AI is more than an aspiration.
For more on the rules shaping South African business, see our Regulatory & Policy coverage.
