With the municipal elections now just days away, South Africans are bracing for a digital storm unlike any before. On 4 November, voters will go to the polls to elect local leaders who will oversee the basic services that shape everyday life, from water and electricity to housing and sanitation. But in the long run-up to that moment, a quieter battle has been unfolding online. The Council for Scientific and Industrial Research (CSIR), together with the Government Communication and Information Systems (GCIS), hosted a virtual roundtable titled “Elections in the era of Artificial Intelligence” with the explicit aim of protecting the integrity of the upcoming vote. The event was attended by a wide range of stakeholders, including media professionals, researchers, government communicators and fact-checkers, all of whom came with the same concern: the rise of AI-generated misinformation has made it more difficult than ever to know what is true and what is manufactured. This is not an abstract debate about the future of technology. It is a matter of democratic survival. If voters cannot trust the content they see on their phones, they cannot make informed decisions. And when distrust spreads, faith in the electoral process itself begins to erode. The roundtable was therefore an urgent intervention, a space where practical solutions could be shared, limitations acknowledged, and new alliances formed before the final days of the campaign. Its underlying message was simple but powerful: elections depend on the free exchange of honest information, and that exchange is now under direct attack. The conversations that unfolded during the event revealed both the scale of the threat and the determination of those working to counter it. No one left the roundtable believing the problem was solved, but there was a shared sense that ignoring it would be far more dangerous. The integrity of the local elections, and of South African democracy itself, may well hinge on how well the country navigates this chaotic and deceptive information landscape.
At the heart of the discussion was a new tool developed by the CSIR, presented by artificial intelligence specialist Avi Moodley. Known as the Informed Fake News Advisor Project, it is designed to filter misleading information on various platforms by combining data collection, natural language processing and human oversight. Moodley walked participants through the process: first, researchers identify the keywords that shape election-related discourse; next, they collect relevant content from social media platforms; finally, they enrich that content using the CSIR’s natural language processing services, together with additional data. The aim is not simply to delete or flag false posts, although that is part of the value, but to understand how misinformation spreads, who is spreading it and what kind of narratives it relies on. Moodley expressed confidence in the analytical systems used by the researchers, describing them as accurate enough to be useful in real time during the election season. Yet his presentation also carried an underlying caution. The project is not a magic shield. It is a response to an ever-evolving adversary. Every time a detection method is developed, those who create disinformation adapt, finding new languages, new platforms, new formats and new tricks. The Informed Fake News Advisor Project demonstrates that the CSIR is not standing still, and it offers a valuable layer of protection, but it also reminds those present that technology alone cannot solve a problem that is fundamentally human. It must be paired with journalists who are willing to question official narratives, with platforms that take responsibility for the content they amplify, and with citizens who pause before sharing a shocking message. The conversation around the project made clear that the defence of truth in the digital age is a continuous process, not a one-time fix, and that everyone has a role to play.
Moodley also took the time to explain why AI has made the current moment uniquely dangerous. “Increasingly sophisticated AI-generated audio and images are making it harder to distinguish authentic content from misleading information, raising concerns about the spread of misinformation and disinformation, particularly during elections,” he said. These are not hypothetical concerns. Over the past few years, the world has witnessed convincingly fake videos of public figures, voice clones that can be used to impersonate politicians and family members, and photographs that appear to capture events that never happened. During election season, this kind of content can be used to suppress voter turnout, damage a candidate’s reputation, deepen ethnic tensions, or encourage violence. What makes it even more frightening is that the technology is becoming cheaper and easier to access. Someone with limited technical skill can now generate a believable fake or alter an existing image in a matter of minutes. Moodley acknowledged that the CSIR’s current methods are strongest when dealing with text-based information. Audio is much harder to analyse, and in South Africa this is a significant challenge because of the enormous range of languages used across the country. A misleading voice note in isiZulu, Afrikaans, Sesotho or any of the country’s other official languages can spread quickly, and fact-checking organisations do not have enough multilingual capacity to listen to every audio clip. This is why newer innovations, capable of searching audio content more extensively, are so urgently needed. Without them, a dangerous gap remains open, one that bad actors can exploit. The answer, Moodley suggested, lies not only in better technology but in partnerships that bring together the CSIR, the media, government communication teams and civil society to create a more resilient information environment.
Practical advice for the public came from Cayley Clifford, deputy editor of Africa Check, who offered guidance on how ordinary social media users can protect themselves from visual disinformation. Her message was urgent because fake images are among the most common and most persuasive forms of election misinformation. They can be designed to show a candidate shaking hands with a controversial figure, an empty rally when in fact thousands attended, or a fabricated election poster carrying the logo of a political party or government institution. Clifford recommended a technique known as reverse image search, which allows users to take a photograph and trace its origin across the internet. This simple tool can reveal whether an image is old, whether it has been altered with editing software, or whether it is being used in a different context from the one being claimed. She also pointed out that fake visuals often have subtle flaws that are visible to anyone who looks carefully. “A real image of people has shadows, and it creates lines of perspective. You should see a point of connection of the lines on an object or going towards the same direction,” she said. This kind of visual awareness does not require specialist training. It simply asks people to look at images with a more critical eye. Clifford highlighted the widespread use of fake posters that include government logos and official-looking signs, noting that these are created precisely to exploit the trust citizens place in official institutions. A poster might look legitimate because it carries the weight of a recognizable emblem, but the logo may be slightly distorted, the layout may be inconsistent, or the message may be foreign to the official position of the institution. By teaching social media users to spot these inconsistencies, Africa Check is trying to build a line of defence that does not rely on professional fact-checkers alone. Every citizen who hesitates before sharing a suspicious image becomes part of a larger effort to slow the spread of disinformation.
The roundtable’s warnings were given an uncomfortable real-world example just days before the election. The official Facebook page of the Msunduzi Municipality shared a post containing incorrect information about how the country’s elections work, specifically suggesting that votes are counted somewhere other than the voting station where they are cast. This is false. In South Africa, votes are counted at the voting station at which they were cast, under the eyes of party agents and independent observers. The mistaken post was all the more damaging because it came from an official government account and used the logo of the Independent Electoral Commission (IEC) without the necessary written authorisation required by the Electoral Commission Act. Mbali Ntuli, a well-known political activist and public figure, was among the first to challenge the municipality on Twitter, drawing attention to the error and the legal problem. She wrote: “This is not a small mistake. In an already highly contested election, incorrect information from official government accounts can cause confusion, fuel suspicion and undermine public confidence in the results. Municipalities have a responsibility to educate voters accurately, particularly with just seven days to go. This post needs to be corrected urgently.” Her rebuke was not merely about one municipality. It spoke to a broader pattern of carelessness and the enormous responsibility placed on those who manage public communication. If the authorities themselves fuel suspicion, they hand weapons to conspiracy theorists who are already eager to claim the election is rigged. Falsehoods from official sources are especially dangerous because they are difficult to retract effectively; the correction rarely travels as far or as fast as the original lie. The incident at Msunduzi Municipality illustrated, in real time, the exact dangers the CSIR roundtable was convened to address. It also served as a reminder that misinformation is not always the work of sophisticated foreign actors or hidden troll farms. Sometimes it is the product of a simple mistake on a local government Facebook page, a mistake that, in a tense electoral atmosphere, can have outsized consequences.
Looking beyond the roundtable, the challenge for South Africa is not only to detect fake news but to create an environment where truth is valued and verified. The discussions painted a picture of an information ecosystem under stress, one in which bad actors can use artificial intelligence to impersonate leaders, create false evidence and exploit social divisions with staggering speed. Yet the conversation also offered reasons for hope. The CSIR is building tools that can identify patterns of misinformation, Africa Check is teaching citizens how to verify images, and ordinary South Africans are increasingly aware that not everything on the internet is trustworthy. The task now is to weave these efforts into a cultural habit. Before sharing a startling claim, people can ask themselves: Does this make sense? Who is the source? What other information confirms this? Looking closely at images, checking the logos, examining the shadows and perspective, and using reverse image search are all acts of democratic self-defence. For public institutions, the lesson is equally clear: accuracy must be prioritised, even in the rush to communicate. A single misleading post from an official account can undermine years of trust-building. The virtual roundtable brought together many of the key players, but the real work must happen in every newsroom, every government communications office and every family group chat. The fight against disinformation is not only about stopping lies; it is about building a citizenry that refuses to be manipulated. As the 4 November elections approach, the hope is that South Africans will not simply consume information passively. They will question, investigate and verify, not because they are cynical but because they understand that their democracy depends on it. In the era of artificial intelligence, the act of sharing is no longer innocent. It carries consequences. It can help build trust or it can tear down the fragile confidence on which free elections rest. The responsibility is shared, and the time to act is now. The tools may be technological, but the solution is deeply human: choose truth, practice care, and protect the right of every citizen to vote in good faith.