MagicCVThe Department of Communications and Digital Technologies has published an 86-page Draft National AI Policy. Public comments close at 16h00 on 10 June 2026. We think the policy is vague, unworkable, and dangerous to the local tools helping South Africans find work. We are asking you to read this, and then take ten minutes to say so in an email.
Where we stand
MagicCV supports AI that helps South Africans. The country has one of the highest youth unemployment rates in the world, and the technology that exists right now, used responsibly, can help close that gap faster than anything government has produced in thirty years.
We are not opposed to regulation. We are opposed to this regulation, as written, because it will not do what it claims to do, and it will damage the people it claims to protect.
To be absolutely clear
If this policy passes with the language it currently contains, MagicCV will not be complying with it. Not the vague definitions, not the undefined licensing regime, not the impact assessments the document cannot describe with enough precision to enforce. If the authors could not be bothered to even write the policy clearly, why should we or anyone who cares about the development of South Africa and its people follow it?
We will continue building tools that help South Africans find work. We will comply with POPIA, the Consumer Protection Act, and every other piece of existing legislation that already covers the actual harms this policy pretends to address. We will support any future framework that is written with genuine care, clear definitions, and real input from the people it will affect.
A policy that cannot be understood cannot be obeyed. We will not pretend otherwise.
What is actually wrong with the document
1. It never defines what AI is
Across all 86 pages, there is no working definition of artificial intelligence that tells anyone whether their product falls inside the policy or outside it. The document describes AI as a research field, cites a few textbooks, and moves on. This is not a small gap. Every provision in the policy, the licensing, the audits, the impact assessments, the proposed AI Regulatory Authority, depends on knowing what counts as AI. The policy does not say.
That means enforcement would be whatever the regulator decides on the day. A company can be in scope because someone says so, and out of scope because someone else says so. That is not a rule. It is discretion wearing a suit.
2. Foreign platforms will ignore it. Local ones will carry the cost.
The big AI services the policy is worried about are hosted in foreign data centres. South African law cannot meaningfully compel them. They will continue operating exactly as they do today. The only actors who will actually feel this policy are small South African operators who cannot afford a compliance team, cannot fight a vague enforcement action in court, and cannot relocate to Mauritius.
This is the same pattern that has played out in the taxi industry, in spaza shop regulation, and in every other case where the state wrote rules it could not implement. The rules end up applying selectively, to whoever is easiest to reach.
3. It proposes institutions the state cannot build
The policy proposes, within three years, a National AI Commission, an AI Ethics Board, an AI Regulatory Authority, an AI Ombudsperson Office, a National AI Safety Institute, and an AI Insurance Superfund modelled on the Road Accident Fund. This is the country that cannot keep Eskom running, staff SARS properly, or run a functional Post Office. Citing the Road Accident Fund as the positive comparison is its own answer to whether this is serious.
4. It will make the job search worse, not better
A policy that loads compliance costs onto local job-matching tools, without touching the foreign platforms, will narrow the options for people who already have the fewest options. If tools like MagicCV cannot operate, the alternative is not a better-regulated MagicCV. The alternative is nothing, and recruiters charging fees you cannot pay.
How to comment
You do not need to be an expert. You do not need a lawyer. You need to be a South African with a view, willing to put it on the record. A short, honest email from a real person carries more weight than a long submission from a consultancy.
Submission details
Email: aipolicy@dcdt.gov.za
Subject line: Draft South Africa National Artificial Intelligence (AI) Policy
Deadline: 10 June 2026, by 16h00
Address to: The Director-General, Department of Communications and Digital Technologies, attention Mr M. Mashologu
Open email to submitUseful points to raise
- Ask for a clear, operational definition of AI before any other provision is finalised. Without it, the rest of the document cannot be fairly applied.
- Ask how the policy intends to reach foreign platforms, and whether it accepts that local operators will carry the compliance cost if it cannot.
- Ask why new institutions are being proposed when the existing ones, POPIA, the Information Regulator, the Consumer Protection Act, already cover most of the actual harms the policy describes.
- Ask what impact has been modelled on youth employment and on small technology businesses in the country.
- Tell them who you are. Student, job-seeker, developer, employer, parent. Specific voices land harder than abstract arguments.
What we support
We want a South African AI framework that works. One with a real definition of what it regulates. One that treats foreign and local actors evenhandedly. One built on institutions that exist and function, not on institutions that have to be invented from scratch. One shaped by the people who actually build and use these tools.
That framework is possible, but it has to be fair, equitable and based in reality. This is not that framework.