Ubiquity AI on the withdrawal of South Africa’s Draft National Artificial Intelligence Policy
South Africa’s AI Policy Debacle Is Not a Setback but Necessary Reckoning
South Africa’s Draft National Artificial Intelligence Policy, approved by Cabinet on 25 March and published in the Government Gazette on 10 April 2026 for public comment, was withdrawn after it emerged that at least six of its 67 academic citations were AI-generated hallucinations fake articles attributed to real journals, and research credited to authors who never wrote them.
The parliamentary portfolio committee chair suggested the department “skip using ChatGPT this time” when redrafting. It would be easy to treat this as dark comedy. We do not. We treat it as a warning that South Africa ignored, and one that Ubiquity AI formally placed on record.
We Said This. In Writing. Both To Government and to Parliament.
As part of the compulsory public participation process under PAJA, Ubiquity AI submitted a formal policy response to the Department of Communications and Digital Technologies. We also engaged Parliament separately about our concerns. In our submissions, we identified a range of technical and sovereign-risk issues that the draft policy failed to address with any meaningful rigour. Among them:
The Double Bias Problem - AI models trained on biased datasets, subsequently re-annotated by human reviewers who introduce a second layer of bias, compounding rather than correcting the original error.
Reinforced Human Training - feedback mechanisms that entrench cultural, linguistic, and socioeconomic assumptions embedded at the point of model creation, often invisible to the end-adopting country.
Foreign Annotators - the overwhelming majority of AI training and annotation work occurs outside Africa, by annotators who do not share the linguistic, cultural, or socioeconomic contexts of South African citizens. The outputs of these models reflect that absence.
We do not view these issues as abstract academic concerns, but sovereign risks. A nation-state that deploys AI systems it did not build, cannot audit, and does not understand has ceded a form of digital sovereignty that is very difficult to reclaim.
The now-withdrawn draft policy was, in our assessment, dangerously light on technical requirements. It proposed governance architecture - commissions, ethics boards, ombudspersons - without first answering the foundational engineering and data questions that would make any such architecture meaningful.
The Harder Lesson
The irony is precise: a policy designed to govern AI was apparently drafted, at least in part, with unverified AI - and it fell into the very same shortfalls it was aiming to address. This is not merely embarrassing. It is instructive.
South Africa is one of approximately 150 countries in the position of an AI importer rather than an AI producer. We do not train the frontier models. We do not control the data pipelines. We do not set the architectural priorities of the systems our government, businesses, and citizens are increasingly reliant upon.
That asymmetry carries real risk - and the speed with which AI adoption is being evangelised, often by those with a direct commercial interest in that adoption, only deepens it.
Regulatory capture is a genuine threat. The loudest voices in any AI policy process are frequently the ones with the most to gain from a permissive framework. South Africa - and the broader cohort of AI-importing nations - must be deliberate about whose interests any given policy actually serves.
Our Position Is Unchanged
AI is not a magic bullet. It is a tool - powerful, consequential, and, as this episode demonstrates, capable of compounding the errors of those who wield it without rigour.
We urge Government, in redrafting, to resist the pressure to move at speed simply to signal ambition. The right policy, delivered late, is infinitely preferable to a flawed policy delivered on a headline-friendly timeline. South Africa has an opportunity - if it chooses to take it - to produce a policy framework that is technically credible, sovereignty-conscious, and genuinely protective of its citizens.
That is the policy Ubiquity AI will continue to advocate for. We remain available to contribute to that process.
Kaveer Beharee
Founder: Ubiquity AI
kaveer@ubiquity-ai.com