South Africa's rewritten national AI policy is now due before cabinet by March 2027, months later than first promised. The officials drafting it have until 30 November to hand it to communications minister Solly Malatsi.

"We are working really under pressure," said Jeanette Morwane, acting deputy director-general at the Department of Communications and Digital Technologies and project manager for the policy. The public will only see the new draft once cabinet approves it for comment.

This is the second attempt. The first draft was published for public comment on 10 April and withdrawn on 26 April, after it emerged that its reference list contained fictitious sources, most likely generated by AI and never checked. At least six of its 67 references did not exist.

In May, Malatsi appointed a seven-member independent expert panel chaired by Wits University AI researcher Benjamin Rosman, and a government drafting team now includes the department, the presidency, the Department of Science, Technology and Innovation, the State Security Agency and the Department of Trade, Industry and Competition.

The timetable Morwane set out is long. The revised policy and a discussion document go to the minister by 30 November. Internal government consultations run from December to February. Cabinet then considers it by March 2027, mainly to approve it for public comment. Only after the comments are processed does it go back to Cabinet for final approval, and no date has been given for that. The deadline has already slipped.

Morwane said discussions are centred on sovereignty over computing power, data and AI models, including ensuring large language models work in South Africa's official languages, and that an implementation road map is being written alongside the policy.

"What's missing for us is there is a huge need for investing in computing infrastructure," she said. One of her slides warned that while South Africa leads its region on AI readiness, regional peers are moving faster on putting AI to use, citing Egypt's national language model, Karnak.

TechCentral's Fanie van Rooyen reported the new timetable from a South Africa-Finland dialogue on AI in Pretoria. TechCentral also covered Malatsi's withdrawal of the first draft, which he said was "not a mere technical issue". CNBC Africa noted that the withdrawn draft had proposed five new oversight bodies, including a National AI Commission and an insurance fund to compensate people harmed by AI.

For businesses, the practical point is that South Africa will not have a final AI policy until well into 2027, and an actual AI law would come later still. That does not mean AI is unregulated in the meantime. Existing laws already apply, as the Information Regulator showed when it fined FNB under POPIA this week, and as we set out in our guide to AI at work.

Financial firms are also waiting on the FSCA, which is holding back on its own AI rules. The shift in focus is the more encouraging part. The first draft leaned heavily towards new oversight bodies. This one talks about computing capacity and models that speak South Africa's languages, which is where the country falls behind, and where local builders, like those who could make data centres here, have a role.

The lesson from April also applies to every business: AI can draft quickly, but a person has to check the facts before anything carries your name.

You might also like our piece on City Power's AI tender, UCT's Google-funded AI research, and Anthropic's AI safety bill.

Get more SA tech and business news and subscribe to The Open Letter.