Within a six-month window, South Africa offered two contrasting demonstrations of how a middle-income state can try to govern Big Tech. In November 2025, the Competition Commission closed a 24-month inquiry that extracted a R688 million (roughly $42 million) support package from Google and concrete behavioural commitments from the largest platforms. In April 2026, the Department of Communications and Digital Technologies published — and then withdrew — a national AI policy that had itself been drafted with an AI tool, complete with academic citations that do not exist.
The juxtaposition is not just embarrassing. It is the clearest case study yet of why the method of tech regulation matters more than its ambition.
The inquiry that did the homework
The Commission's Media and Digital Platforms Market Inquiry (MDPMI) released its final report on 13 November 2025. It was the product of five rounds of information requests, public and in-camera hearings, a consumer survey, and a provisional-report comment process. Its central finding was specific and evidentiary: Google's search dominance, combined with AI-generated answers, has sharply cut referral traffic to South African newsrooms, while the platform pays nothing for the news content it surfaces. The Commission also found Google's ranking favoured foreign outlets over local and vernacular ones.
The remedies tracked the harm. Google and YouTube agreed to a R688 million package covering content licensing, innovation grants, and capacity-building, per the African Antitrust summary of the settlement. Google committed to stop self-preferencing its own properties and to surface local news. Crucially, the Commission required AI firms to give South African publishers the same content controls and opt-out options already available in the EU, plus biannual training — importing a proven standard rather than inventing an untested one.
It is worth steelmanning the platforms' objection. There is a real risk that bargaining codes and content-payment mandates entrench incumbents, raise costs that get passed to users, and treat a structural shift in how people find information as if it were simple expropriation. Australia's News Media Bargaining Code has been criticised on exactly these grounds. But the South African remedies are narrower than the maximalist version: they target a documented decline in referral traffic and a specific refusal to license, they borrow the EU's existing opt-out architecture, and they were negotiated rather than imposed by fiat. That is proportionate regulation — anchored to evidence, calibrated to a defined market failure, and stopping short of a blanket levy.
Notably, the Commission declined to impose financial remedies on Meta, X, and TikTok, recommending instead self-regulation frameworks and an independent social-media ombudsman. Restraint where the evidence was thinner is itself a sign of disciplined process.
The policy that skipped the homework
The AI policy went the other way. Cabinet approved the draft in late March 2026; it was gazetted for public comment on 10 April 2026. Within roughly two weeks, News24 reported that at least 6 of its 67 academic citations were fabricated — attributed to real scholars but never published. Editors at the South African Journal of Philosophy, AI & Society, and the Journal of Ethics and Social Philosophy confirmed the cited articles never appeared in their pages, as The Next Web documented. On 27 April, Minister Solly Malatsi withdrew the policy, calling the fictitious sources an "unacceptable lapse" that "compromised the integrity and credibility of the draft" and proof that "vigilant human oversight over the use of artificial intelligence is critical."
The strongest case for an ambitious national AI policy is genuine. Africa's largest economies are, by their own admission, dependent on Google, Microsoft, Nvidia, and Meta for compute, capital, and expertise; a coherent strategy is a legitimate response to that dependence. But ambition is not the variable that failed here. Basic verification did. A document meant to set the standard for how the state governs AI was produced without anyone checking whether its own evidence base was real.
Sovereignty rhetoric meets institutional capacity
The contrast lands inside a continent-wide debate. As Rest of World reported, South Africa, Nigeria, Egypt, and Kenya have all drafted AI strategies naming dependence on U.S. tech firms as a strategic threat, and proposals for a $60 billion African AI fund and a continental AI council are circulating. South Africa reached that same conclusion in the very draft it had to retract.
The lesson is not that AI sovereignty is a bad goal, nor that competition enforcement is the only tool. It is that the two episodes ranked the same government's capabilities. The Commission won real concessions because it built an evidentiary record over two years and pegged its demands to standards that already work elsewhere. The AI policy collapsed because it substituted generative output for that work.
For a publication that favours innovation and proportionate rules, the takeaway is uncomfortable for techno-optimists and statists alike. Governments racing to regulate AI should be held to the verification standard they intend to impose on everyone else. And the most effective check on platform power in South Africa came not from a grand strategy document but from a competition regulator that did the unglamorous, evidence-heavy work. Capacity, not ambition, is the scarce resource — and it is the thing worth funding first.