Democracy has always depended on a shared sense of reality. Citizens need reliable information to decide who to trust, which policies to support, and where to cast their votes. But in the age of artificial intelligence, that shared reality is becoming harder to find. As South Africa prepares for municipal elections, in which some 29.2 million registered voters will choose local government leadership, a virtual media roundtable hosted by the Council for Scientific and Industrial Research (CSIR), in partnership with the Government Communication and Information System and other stakeholders, set out to confront a growing challenge: digital narratives and AI-driven disinformation. The message from researchers and experts was clear: low-cost AI tools now allow malicious actors to deceive large numbers of people using highly accurate synthetic media. What used to require sophisticated production studios or state-backed propaganda machines can now be done with an ordinary smartphone and a free app. The roundtable examined how fact-checking, public communication, science, technology and collaboration can help safeguard the integrity of public information. But beyond the official discussions, the underlying concern was deeply human. Imagine opening your phone and seeing a video of a beloved leader saying something they never said, or hearing a voice note that appears to expose a secret that never existed. In the past, such content would have been dismissed as obvious forgery. Today, it can look and sound real enough to change minds, stoke anger, and alter the course of an election. This is not a hypothetical threat; it is already unfolding in South Africa and across the continent. The stakes are enormous, because if voters no longer know what to believe, democracy itself begins to dissolve.
Tom Gshlain, a digital researcher at the Centre for Analytics and Behavioural Change (CABC), painted a stark picture during the roundtable. The CABC has been monitoring social media platforms such as Facebook, X and TikTok ahead of the elections, and what they have found is worrying: AI-generated videos, AI-generated posts and even AI-generated score cards of political parties are circulating widely. This content is designed to look organic, to fit seamlessly into the feeds of ordinary South Africans, and to provoke emotional responses. Gshlain noted that this is not a uniquely South African phenomenon. AI-generated content was shared during Ghana’s 2024 elections and in Nigeria’s 2023 elections, and similar tactics are likely to appear wherever democratic contests take place. The danger lies in the speed and scale of AI. It allows for the rapid creation of credible-looking false stories involving text, images, video and audio, and these can outpace traditional reporting methods. Gshlain offered a striking example: an AI video designed to look like the late president Nelson Mandela advising voters against the African National Congress. There have also been fake videos involving former US presidents Donald Trump and Joe Biden. These are not crude edits; they are carefully constructed pieces of synthetic media designed to deceive. Worse, the barrier to entry is almost nonexistent. The CABC has identified six commonly used AI tools, and Gshlain made the point memorably: making anyone say anything can cost less than a mobile data bundle. With a phone and a short recording, anyone can now create a convincing video or clone a voice. Most of these tools are free to try. That means a teenager, a political operative, or a foreign actor can produce convincing disinformation without a studio budget. These tools can fake an endorsement, manufacture a leaked voice note in a WhatsApp group, or create a fake poll or election result styled as a television news broadcast.
There is also a more insidious twist, one that makes all of us more vulnerable to manipulation. Once people know that fake news exists, anyone caught on a real clip can simply claim that the clip was generated by AI. Gshlain called out this dynamic: “Once people know fake news exists, anyone caught on a clip can say it was AI.” This so-called “liar’s dividend” allows public figures to evade accountability for genuine statements, and it makes it even harder for ordinary citizens to sort fact from fiction. A legitimate recording can be dismissed as a deepfake; a fake one can be accepted as truth. In such an environment, facts no longer travel with certainty, and trust becomes a casualty. The CABC has developed practical countermeasures. It monitors for cloned voices or fake posts, verifies sources through cross-checking, and engages directly with harmful content using facts. This is essential work, but it is not enough on its own. The roundtable also highlighted the role of fact-checking and public communication in countering misinformation and disinformation. Fact-checkers are on the front lines, but they are often racing against algorithms that amplify outrage faster than corrections can reach the same audience. Public communication efforts must therefore be proactive, not reactive. Governments and institutions need to provide clear, trustworthy information before the void is filled by lies. The challenge is not simply to debunk false content after it has spread; it is to create an information environment where false content struggles to gain traction in the first place. This requires media literacy, community engagement, and a willingness to address the emotional and social drivers that make people share disinformation. It also requires a shared commitment to supporting journalists, researchers and civil society organizations who are trying to protect the public sphere.
The CSIR is stepping up to meet this challenge, but it is calling on others to join. Avi Moodley, acting research group leader for natural language processing and principal AI specialist at the CSIR, made a passionate case for collaboration. He said that organizations tackling misinformation and fake news cannot afford to work in isolation. “The challenge is enormous. Whether it is data, infrastructure, resources or funding, all these elements can make a significant difference to our coordinated efforts. We would welcome your participation in strengthening and expanding this network.” Moodley also spoke about the overwhelming volume of content in today’s information ecosystem. It is simply impossible for ordinary people to determine whether everything they receive can be trusted. “You do not know what is real and what is AI-generated. If we, as researchers, experience this fatigue when analysing information and data, can you imagine what ordinary South Africans are experiencing when they are overwhelmed with information and are uncertain about what is real or fake?” This is a crucial point. Researchers are trained to analyse sources, verify claims and detect patterns, and even they feel exhausted by the flood of synthetic content. For the average voter, the situation is far more confusing. Advances in AI have, to some extent, industrialised the spread of misinformation, disinformation and malinformation. AI can be used to create synthetic content, develop and amplify emotionally charged narratives, and produce material that could potentially manipulate perceptions in specific contexts. The flooding of the information ecosystem with large volumes of content has created an environment unlike anything seen in previous elections. Moodley warned that future elections will face even greater challenges, and the response must be coordinated. No single organization, no matter how sophisticated, can solve this problem alone. Data sharing, joint research, pooled infrastructure and shared funding can make the difference between a resilient democracy and one that is overwhelmed by lies.
The Electoral Commission of South Africa has already recognized the urgency of the moment. Earlier this month, the IEC announced a partnership with Google, Meta, Microsoft, TikTok and the non-profit organization Moxii Africa, formerly Media Monitoring Africa, to ensure the integrity of election information and counter disinformation. This is not the first time such a framework has been used; it is a sequel to a similar agreement signed with the platforms ahead of the 2024 national and provincial elections. But the timing could not be more important. With local government elections approaching, the online space is about to become even more crowded with political messages, rumours and attacks. Chief electoral officer Sy Mamabolo explained why the partnership matters. He said that the inclusion of the online platforms gives the commission direct access to the platforms and a real fighting chance against disinformation. “A technically-flawless election can still lose public trust if citizens are overwhelmed by disinformation. Therefore, the partnership eliminates barriers to direct access to the platforms and enables instantaneous remedial action against disinformation peddlers,” Mamabolo said. These words capture a profound truth about modern elections. A ballot can be perfectly printed, polling stations can run smoothly, and votes can be counted with complete accuracy, but if the public is drowned in lies, the legitimacy of the entire process can be called into question. Disinformation does not need to change the outcome to damage democracy; it only needs to make people doubt the outcome. By working directly with technology companies, the IEC hopes to remove suspicious content quickly, correct false narratives, and ensure that official information reaches voters through the same platforms where they are already spending their time. But direct access is only part of the solution. The partnership must also involve transparency, accountability and a shared commitment to protecting the public interest over political advantage.
Ultimately, technology alone will not save democracy. The CSIR roundtable underscored the need for collaboration among scientists, government, tech companies, media, civil society and citizens. Science and technology can help detect synthetic media, track narratives and provide fact-checking tools. But human judgment and media literacy are equally critical. As voters, we must pause before sharing something that makes us angry or afraid. We must ask where it came from, whether the source is credible, and whether we have seen any confirmation from trusted news outlets. We must be willing to admit that a video or a voice note is not necessarily proof of anything. At the same time, public institutions must be transparent, quick and clear in their communication. They cannot wait for a lie to go viral before responding; they need to be present, online and credible before the next crisis erupts. The battle against AI-driven disinformation is winnable, but only if we treat it as a shared responsibility. It is not enough for researchers to publish reports or for fact-checkers to issue corrections. Every person who receives a suspicious WhatsApp message has a choice: to spread it or to stop it. Every platform that profits from engagement has a responsibility to slow the spread of synthetic content. Every political party that benefits from false narratives has a choice to disavow them or exploit them. In the end, the defence of democratic information is not just a technical challenge. It is a moral one. The CSIR, the CABC, the IEC and their partners are building a network of resilience, but they cannot do it alone. As Mamabolo warned, a technically flawless election can still fail if trust is destroyed. So the task is not only to protect ballots, but to protect the shared reality that makes democracy meaningful. That responsibility belongs to all of us.