On the eve of South Africa’s local government elections, a quiet but critical battle is being fought over the truth itself. The usual campaign noise has been joined by something more sinister: an invisible digital storm of manipulated images, cloned voices, and algorithmically weaponised rumours. Public officials and independent fact-checkers recently gathered around a table to confront this threat, and their message was both urgent and deeply human. This is not just a story about algorithms and artificial intelligence; it is a story about whether voters can still trust what they see, hear, and read before they cast their ballots. The institutions responsible for protecting the democratic process are no longer dealing only with printed pamphlets or televised debates. They are trying to navigate a world where content is everywhere, where social media feeds are shaped by machines, and where a voter in a remote village might receive a hoax in a WhatsApp voice note at the very moment they are deciding how to vote. For Sandile Nene of the Government Communication and Information System, this is not an abstract policy headache. It is a constitutional and socioeconomic obligation. South Africa, he argued, is living in an “attention economy,” where political reality can be drowned out by endless, algorithmically amplified noise. And cutting through that noise is about more than information management. In his words, there is always a direct link between an informed society and economic development, a principle rooted in the World Summit on the Information Society and South Africa’s own National Development Plan. Put simply, a citizen who cannot access reliable information cannot fully participate in democracy, and a democracy deprived of informed citizens cannot build a fair economy.
The challenges, however, are enormous, and they are growing in ways that are difficult for older institutions to absorb. Public trust in governments and the media has been fraying for years, and this trust deficit has opened the door to unethical uses of artificial intelligence. For South Africa, the stakes are particularly high because the rewards of getting AI right are extraordinary. Global research from Accenture has suggested that the country stands to gain around R1.3-trillion if it can implement ethical AI frameworks successfully. But Nene was careful to warn that the same technology, left unregulated and used with destructive intent, presents an immediate danger to the country’s social fabric. He highlighted a stark and uncomfortable reality: nearly 29 million South Africans live at the “bottom of the pyramid,” a category government researchers describe as information-deprived. These are citizens with little access to alternative news sources, few digital verification tools, and often no reliable way to tell the difference between an authentic government notice and a synthetic piece of propaganda designed to incite fear or anger. To respond to this, the Government Communication and Information System has built partnerships with the Council for Scientific and Industrial Research and with major international digital platforms, focusing on content moderation and takedown strategies. It has also leaned heavily on its partnership with the Electoral Commission of South Africa to protect institutional legitimacy. But even with these measures, Nene warned about the “unspoken harm” of peer-to-peer sharing. A doctored video or a fake victory declaration can spread organically through WhatsApp and Telegram long after it has been formally tagged or removed. Crucially, he reminded citizens that under South African domestic law, individuals can be held personally accountable for sharing or liking harmful content that incites violence or causes mass confusion. The accountability, in other words, no longer rests only with the original creator; it rests with every single thumb that taps forward.
While government officials are working on systemic policy responses, independent researchers are monitoring the battle in real time, and what they are seeing is both more subtle and more troubling than people usually imagine. Researchers from Africa Check and the Centre for Analytic Behavioural Change brought a grounded, tactical perspective to the roundtable. Cayley Clifford, a researcher at Africa Check, offered a surprising nuance: for all the fear surrounding advanced, Hollywood-style deepfakes, South Africa’s disinformation ecosystem is thriving mostly through very basic techniques. Most misleading content relies on “false context,” which is simply a genuine, unedited image or video stripped of its original setting and repositioned to tell a completely different story. A clip from an old protest in one country can be presented as a current event in another; a random video from years ago can be made to look like evidence of election fraud today. This is not high-tech wizardry; it is old-fashioned deception with a modern distribution network. Clifford also drew a sharp distinction between two tiers of AI-generated manipulation. The first tier, the truly convincing deepfake, remains relatively rare because it demands advanced technical expertise and expensive computational resources. The second tier, however, is everywhere. These are clumsy fabrications made quickly with free online software. They often have poor lip-syncing, distorted hands or facial features, and a generic, synthetic look. They would not fool a careful analyst, but they do not need to. In the fast-moving chaos of a typical social media feed, they can easily persuade a busy, anxious, or exhausted voter. In that sense, the danger is not that disinformation is becoming harder to spot; it is that there is simply so much of it, moving so fast, and preying on the emotions of people who are already scared and uncertain.
Tom Gshlain, a researcher and dialogue facilitator at the Centre for Analytic Behavioural Change, pushed the conversation further into the disturbing realm of speed and affordability. The primary danger of contemporary AI, he argued, is not necessarily its perfection but its immense scale. It allows harmful, polarising narratives to be generated and weaponised faster than fact-checkers can fact-check them, faster than platforms can remove them, and faster than the human mind can process them. His team’s social monitoring across X, Facebook, and TikTok has observed a worrying democratic shift in text-to-video manipulation. In earlier years, creating a fake video or a cloned voice required media studios, specialised software, and substantial budgets. Today, the financial barriers have all but collapsed. Voice-cloning tools like Resemble AI can deliver up to 95% accuracy for as little as R33 per month. Specialised video generation suites are accessible for roughly R115 per month. That means an entire political disinformation campaign can be run for less than the cost of a standard internet data bundle. The implications are chilling. A malicious actor with a grudge, a tiny budget, and a willingness to harm democratic trust can now manufacture convincing audio clips of a mayoral candidate saying things they never said, or synthetic videos of electoral officials acting corruptly, and distribute them to thousands of voters in a single evening. The inequality of this moment is deeply troubling: professional fact-checkers are overworked and underfunded, while the tools of deception are becoming cheaper, faster, and more accessible to anyone with an internet connection and bad intentions.
As election day approaches, both public officials and independent researchers agree on a humbling truth: state policy, platform regulation, and automated content takedowns are only half the battle. The final line of defence rests with the individual voter, but not in a way that blames ordinary people for being deceived. Instead, it is an invitation to build a culture of digital discipline and collective care. The experts were clear that many disinformation campaigns work not by convincing people through rational argument but by triggering immediate, intense emotional responses. A graphic image, a shocking headline, a sensational audio clip, a voice note claiming that a polling station has moved at the last minute: these are designed to make the heart race, to make someone share before thinking. The advice from the roundtable was gentle but serious. If you feel an overwhelming surge of anger, fear, or triumphant outrage, pause. Do not share immediately. Take a moment to check the internal clues: does the video look slightly off? Do the voices sound natural? Are the faces distorted? Then check external clues: can you find the story anywhere else? Does any reliable media outlet confirm it? Voters are encouraged to practice media verification skills on educational platforms, to familiarise themselves with how images and audio can be manipulated, and to lean on trusted sources rather than an emotional impulse. But above all, the golden rule remains simple and powerful: if you are ever in doubt, do not pass it on. This is not merely a personal choice; it is a civic obligation. Under South African law, sharing harmful content can carry legal consequences, but beyond the legal threat, there is something more profound. Every time an individual chooses to pause and verify rather than share and pray, they are protecting the entire democratic culture around them.
In the end, this is not just a story about artificial intelligence or social media regulation. It is a story about the fragile, human architecture of trust. South Africa has made remarkable strides in building electoral institutions, independent media, and civil society organisations that protect the integrity of public life. But in the digital age, trust cannot be manufactured solely by institutions; it must be practised by people. The arrival of AI has changed the rules of the game. It has made it incredibly easy to manufacture believable lies, to exploit fear, and to erode confidence in democratic processes. Yet the response from those gathered at the roundtable was not one of despair but of determined realism. Yes, the threat is serious. Yes, millions of South Africans are vulnerable. Yes, the tools of deception are cheap and widespread. But honesty is still powerful. Verification is still a radical act. The truth, once recovered, still carries weight in human hearts. What is needed now is a collective awakening: among government leaders, who must invest in ethical AI frameworks and public literacy; among technology companies, who must take responsibility for their algorithms; among fact-checkers and civil society, who must continue to chase the lies with courage; and among ordinary citizens, who must learn to withstand the emotional pressure of a manipulated feed. There is something deeply human about pausing before sharing, about asking questions, about demanding evidence. It is the same impulse that makes democracy possible in the first place: the belief that people are capable of reason, that they can govern themselves, and that the truth, however imperfectly pursued, is worth defending. As South Africa prepares to vote, that belief is itself on the ballot. And with it, the quiet promise that no algorithm will be allowed to speak louder than the people.