South Africa has just run a live experiment in disinformation governance. It did so without a platform-accountability statute, and the results of that experiment bear on the legislative debate that follows.
What the study says
On 15 October 2026, at the Media Freedom Festival in Johannesburg, the Konrad Adenauer Foundation's Sub-Saharan Africa media programme launched A South African Legal Framework to Govern Online Harm: What Might We Learn from the EU's Digital Services Act?, written by Nicole Fritz and Justine Limpitlaw. According to the KAS launch page, the authors find a gap between existing laws that touch aspects of online harm and a coherent system of platform accountability. They propose obligations scaled to a service's size, reach and influence, with the largest platforms carrying duties to assess systemic risks. Those risks include disinformation, election interference and hate speech. The study also proposes researcher access to platform data and ICASA as lead regulator.
The page also records a caution that is easy to miss: the authors warn against rules that encourage excessive content removal. They favour clear complaint procedures, meaningful appeals and explanations for moderation decisions, and place freedom of expression at the centre of their recommendations. I could read only the launch page, not the full study text, so this analysis rests on that summary.
The strongest case for a statute
The case for legislation is serious. Platforms decide, at scale and in private, what South African voters see. Without a statute, voluntary cooperation lasts only as long as each company's priorities do. Researchers cannot audit what they cannot access, and users have no guaranteed route to challenge a removal. The EU's Digital Services Act was written to address exactly these asymmetries, and a country without such rules depends on goodwill.
What the election actually tested
The Independent Electoral Commission's approach shows a different model. The Disinformation Code was gazetted on 4 September 2026 under section 88(a) of the Local Government: Municipal Electoral Act 27 of 2000. It is part of the Electoral Code of Conduct and applies from publication until the declaration of results.
Its duties fall on parties, candidates and anyone acting for them. They must publicly retract and correct false or misleading information within 36 hours of becoming aware of it, and report online disinformation to the Real411 platform within the same window. They must also label AI-generated content and political advertising and avoid fake accounts and bot networks. Breaches attract the same sanctions as other breaches of the Electoral Code of Conduct.
The code contains no takedown or platform-liability obligations. The SAnews summary shows no provision compelling any intermediary to remove anything. The pressure lands on the people with the most direct control over the message, which is a defensible place to put it.
The platform side was handled by agreement. On 20 October 2026 the IEC announced a partnership with Google, Meta, Microsoft, TikTok and the non-profit Moxii Africa. The partners committed to cooperating in good faith, respecting existing law with no requirement to share confidential user data, and running a working group and training for parties. The announcement also left platforms room to apply their own policies, including removal, advisory warnings and delisting. It describes no binding duties beyond these commitments.
Real411, run by Moxii Africa, is a referral system rather than a censor. Its election page says it submits independent outcomes to the IEC, which decides what measures to take. The page does not describe takedown orders.
Why the sequencing matters
This architecture avoids the problem that most worries free-expression advocates about notice-and-takedown regimes. When a legal duty to remove content attaches to an intermediary on a deadline, the cheapest compliance strategy is to remove first and ask questions later. The 36-hour clock here runs against a party's own correction duty, not a platform's removal duty. A candidate who must correct a false claim does not silence anyone else.
The design is not free of risk. Real411 and the IEC adjudicate political speech during a campaign, and contested calls will arise. The independent-outcome model still matters, because it keeps a visible, reviewable step between a complaint and any consequence. A permanent statute should preserve that step rather than replace it with a deadline.
The underlying problem is real. The Institute for Security Studies reported on 24 September 2026, citing the IEC Voter Participation Survey, that self-described confidence in recognising fake news fell from 62% in 2023 to 52% in 2025/26. Only 18% of respondents say they regularly verify election information before sharing it. It also noted that social media as a news source rose from 8% in 2021 to 23%. These figures argue for media literacy, rapid correction and transparency, which are tools that do not require removing speech.
What a permanent law should borrow
The study's best ideas are the ones that add information rather than subtract speech: transparency reporting, vetted researcher access, statements of reasons, and meaningful appeals. These raise accountability without creating an incentive to over-remove. Proportionality by platform size is also sound, since a regime built for the largest services should not crush local startups.
Three cautions follow from the election period.
- Keep election rules election-specific. The code's sanctions work because they are tied to a defined campaign and an existing legal framework. Extending a 36-hour standard to ordinary political speech year-round would be a different and much riskier measure.
- Do not import a takedown clock without the safeguards. If notice-and-action duties are adopted, they should come paired with reasons, appeals and penalties for abusive notices, as the study itself urges.
- Fund the regulator before expanding its remit. The KAS summary stresses that adequate funding, investigative powers and meaningful penalties would be essential to ICASA's role. A thinly resourced regulator facing the largest global platforms will default to the easiest lever, which is removal requests.
South Africa's cooperative model has limits. It rests on goodwill, and the platforms' own announcement stresses that they will keep using existing policies. The right response is to codify transparency and process, not to legislate speed of removal. The election code shows that responsibility for correcting falsehoods can be placed on those who spread them, with platforms as partners rather than deputised censors. A statute should build on that finding.