Mr Elokusa
Article

I Commented on South Africa's Draft National AI Policy. Here's What I Said

On 10 April 2026, Government Gazette No. 54477 opened the country's Draft National AI Policy for public comment. I sent one in. Here's a short read of what the policy says, and the two things I told the department they need to grapple with more honestly.

The Gazette, in Plain English


Government Gazette No. 54477 dropped on 10 April 2026 and inside it is Notice 3880: the Draft South Africa National Artificial Intelligence Policy. Minister Solly Malatsi (Communications and Digital Technologies) signed the notice. The draft itself was approved by Cabinet on 25 March 2026, and the foreword is signed by Acting Director-General Omega Shelembe.

The public has 60 days to submit comments. The window closes 10 June 2026 at 16h00. Submissions go to Mr M. Mashologu (Deputy Director-General: Digital Society and Economy) at aipolicy@dcdt.gov.za, with "Draft South Africa National Artificial Intelligence (AI) Policy" in the subject line. Late submissions may not be considered.

The document is 86 pages. The drafters openly describe it as "a work-in-progress" and "a point of departure" — they've deliberately left some language open-ended because they want industry and subject-matter experts to help sharpen it during this comment window. That is the whole reason they're asking.




What the Policy Is Trying to Do


A few threads run through the document:

Staged implementation over three years.
Year 1 (2025/26): finalise the policy, identify draft regulatory requirements for unacceptable risks, start drafting the National AI Policy Guidelines. Year 2 (2026/27): publish the guidelines, implement regulations for high-risk use cases, draft regs for medium/low-risk cases, develop sectoral AI strategies, secure funding. Year 3 (2027/28): full rollout, likely updated to match whatever AI looks like by then.

Ethics and governance.
The policy proposes an AI Ombudsperson and an AI Ethics Board. Heavy focus on fairness, bias mitigation, data sovereignty, and alignment with the Constitution and Bill of Rights.

Capacity and infrastructure.
Integrating AI into primary, secondary, and tertiary education. Building AI hubs and supercomputing facilities aimed at local startups and SMEs.

Situated in existing law.
It pulls on POPIA, the Copyright Act, the Electronic Communications Act, PAIA, and the Cybercrimes Act, and interfaces with the AU Digital Transformation Strategy, the OECD AI Principles, and UNESCO's Recommendation on the Ethics of AI.

A "whole-of-government" posture.
The policy sets national norms, but sector-specific working groups (manufacturing, energy, health, finance, agriculture, etc.) will do the actual implementation.

The spirit is good. The ambition is obvious. My concern is with two specific gaps.




What I Submitted


I sent the Department the note below. I work as an AI & Automation Engineer at Silvertree Brands, a South African ecommerce company, where I build and deploy production AI across logistics, customer service, and merchandising. That's the vantage point the comment is written from.

Dear Mr Mashologu and the DCDT Policy Team,
>
I write in response to the invitation for public comment on the Draft South Africa National Artificial Intelligence (AI) Policy, published in Government Gazette No. 54477 on 10 April 2026.
>
I work as an AI and Automation Engineer at Silvertree Brands, a South African ecommerce company, where I build and deploy production AI systems across logistics, customer service, and merchandising functions. I offer the following practical observations for consideration.
>
On data sovereignty and implementation reality
>
The policy rightly identifies data sovereignty as a concern. In practice, this tension is immediate. The most reliable path to deploying functional AI features today (chatbots, recommendation systems, document processing) runs through the APIs of companies that own and operate the underlying models, primarily based in the United States. Using them means South African business data transits foreign, privately owned infrastructure. There is currently no local alternative of comparable capability. The policy should grapple honestly with this tradeoff rather than assume sovereignty and capability can be achieved simultaneously in the short term.
>
On AI adoption patterns in the private sector
>
From what I observe, AI adoption in South African businesses is either structured and intentional, or entirely absent. There is very little middle ground. The policy's focus on upskilling is necessary, but the more urgent gap is at the organisational leadership level. The critical need is for decision-makers who understand enough about AI to commission it properly, define problems clearly, and set realistic expectations. Without this, even well-resourced organisations will continue to invest haphazardly or not at all.
>
I hope these observations are useful to the policy team.
>
Kind regards,
Elokusa Zondi




The Short Version of the Argument


One. The policy wants data sovereignty. It also wants South African businesses to deploy real, working AI. Right now those two things pull in opposite directions, and the document doesn't quite admit it. The models that actually work in production today are hosted by a small number of US companies. If a Cape Town logistics firm wants a working AI feature this quarter, their data is leaving the country. That's not a choice anyone is making out of carelessness — it is the state of the market. A policy that speaks about sovereignty without saying that out loud is harder to act on.

Two. Upskilling is necessary but not sufficient. The interesting thing I see from inside the private sector is that businesses either do AI properly — with intent, budget, and a clear problem — or they don't touch it at all. There is almost no middle category of "we're experimenting cautiously." The reason isn't talent; it's leadership. The bottleneck is decision-makers who understand enough to commission AI well: to scope a problem, buy the right thing, and hold realistic expectations. No amount of training developers fixes that layer.




Why This Matters Now


You have until 10 June 2026 at 16h00 to send a comment to aipolicy@dcdt.gov.za. The policy is explicitly asking for this. It's one of the rare points in the process where the text is still pliable. If you build, deploy, buy, or regulate AI in South Africa — or if you just have a view — this is the moment to put it in writing.

The gazette is public at gpwonline.co.za and the draft policy is available at dcdt.gov.za and gov.za.

Discussion

(0)

Join the first cohort. Unlock badges.

Sign In to Comment