AI Propaganda Machine Exposed in Kenya as Anthropic Investigation Reveals Posts Praising Opiyo Wandayi
By Milton Were Kenya’s political propaganda industry appears to have entered a new and potentially dangerous era after an investigation by US artificial intelligence company Anthropic uncovered the use of AI to manufacture political social media posts, including content praising Energy Cabinet Secretary Opiyo Wandayi. The operation provides a disturbing glimpse into how artificial intelligence can be deployed to manufacture the appearance of public support around politicians and government policies, while simultaneously pushing negative narratives against political opponents. Anthropic’s investigation established that an operator used its Claude artificial intelligence system to generate large batches of Kenyan political posts. The messages were deliberately written in different styles so that coordinated political communication could appear to social media users as spontaneous opinions coming from ordinary Kenyans. Wandayi emerged as one of the political figures who benefited from positive messaging generated through the operation. Other content generated through the same system targeted opposition figures and promoted narratives questioning the strength and unity of President William Ruto’s political opponents. The revelation comes at a sensitive time as Kenya moves closer to the 2027 General Election, when social media is expected to become one of the most important battlegrounds for shaping political opinion. Opiyo Wandayi Praised Through Coordinated AI Posts A significant part of the operation focused on promoting Wandayi and government-related messaging. The Energy Cabinet Secretary was praised over the handling of electricity tariffs, with messages presenting government intervention in positive terms. Hashtags including #PowerReliefKE and #PoweringTheNewKenya were incorporated into the campaign. But the most significant aspect was not simply that Wandayi received positive publicity. Politicians are praised online every day, while supporters have every right to defend government officials they believe are performing well. What makes this operation different is the method used to create the impression that numerous people were independently expressing similar positive opinions. Artificial intelligence was instructed to produce variations of centrally determined messages, effectively turning one political talking point into numerous apparently independent social media posts. That creates the appearance of grassroots enthusiasm without necessarily revealing the coordination behind it. AI Was Asked to Produce 50 Posts at a Time The scale and efficiency of the operation demonstrate why artificial intelligence could fundamentally transform political propaganda. The operator used Claude to produce batches of exactly 50 posts around particular narratives. Instead of manually writing dozens of tweets, a political communications operator could provide the AI with the desired talking points, hashtags and themes and have the system generate numerous variations almost immediately. Existing messages were also fed into the system to be rewritten and made more natural. The objective was straightforward: the messages should not look as though one person or organisation had written them. AI therefore wasn't necessarily deciding the political agenda. A human operator selected the politician, political opponent, issue and desired narrative. The machine's role was to manufacture scale and variety. One person could suddenly sound like 50 people. With multiple accounts and repeated batches, that capability could theoretically be expanded much further. Same Operation Turned Its Guns on Opposition Figures The operation wasn't limited to praising government figures. The same infrastructure was used to promote narratives portraying Kenya's opposition as divided and politically weak ahead of 2027. Former Deputy President Rigathi Gachagua and former President Uhuru Kenyatta were among the political personalities appearing in content produced around opposition politics. This combination makes the operation particularly interesting. Positive narratives were generated around a senior member of Ruto's Cabinet while other messages sought to reinforce perceptions of divisions among political forces opposed to the administration. The political objective of such communication is easy to understand: strengthen favourable perceptions of one side while weakening confidence in the other. What AI changes is the speed and scale at which that can happen. Manufactured Public Opinion Could Become Kenya's Next Political Weapon Kenya is already familiar with coordinated hashtags and political influencers. Political communication teams have for years mobilised online supporters around specific hashtags, talking points and personalities. Artificial intelligence takes that model several steps further. Previously, campaign organisers needed many people to create sufficiently different posts if they wanted an online campaign to appear organic. Repeating identical messages made coordination relatively easy to identify. Generative AI can now rewrite essentially the same political argument dozens or hundreds of different ways. One post can sound angry. Another can sound analytical. Another can sound humorous. Another can appear to come from a frustrated taxpayer. Another can sound like a satisfied voter. Yet all five can originate from exactly the same political instruction. That is the real significance of what Anthropic uncovered. When Propaganda Starts Looking Like Your Neighbour's Opinion The danger is not simply that politicians can generate more content. The bigger problem is that AI can blur the distinction between political advertising and genuine public opinion. Traditional political advertising is relatively straightforward. A politician purchases an advert, their photograph appears on it and voters understand that someone is attempting to persuade them. Coordinated AI messaging can work differently. A Kenyan opens X and sees one account praising a Cabinet Secretary. A few minutes later another account makes a similar argument using completely different words. Then another account repeats the same conclusion. A hashtag begins gaining traction. Soon the user can reasonably begin wondering whether the country is experiencing a genuine change in public sentiment. That perception itself has political value. People naturally pay attention to what they believe other people think. Manufacturing the appearance of consensus can therefore become a powerful political weapon. The Same Technology Was Used for Commercial Marketing Another revealing aspect of the operation was the overlap between political propaganda and ordinary commercial marketing. The same AI-driven techniques were also used to generate promotional content for commercial brands. That suggests an increasingly blurred line between digital marketing agencies, influencer networks and political communications operations. The same operator capable of promoting a consumer product can potentially use virtually identical infrastructure to promote a Cabinet Secretary or attack a political opponent. Only the talking points change. The technology remains the same. That creates the possibility of an entire political communications industry where influence can be purchased and delivered at scale while appearing organic to the ordinary social media user. Who Was Behind the Wandayi Campaign? This is now one of the biggest unanswered questions. Anthropic established how its technology was being used, but the findings did not establish that Wandayi personally commissioned or knew about the operation. Neither did the investigation establish that the Kenyan government ordered the campaign. That distinction is important. Being praised by a coordinated online operation does not mean the person being praised organised it. The real question therefore becomes: who wanted Wandayi promoted? Who selected the electricity tariff issue? Who supplied the talking points? Who decided which hashtags should accompany the messages? Who controlled the accounts that were expected to publish them? Was someone paying for the campaign? And who stood to benefit politically from simultaneously promoting government figures and undermining opposition personalities? Those are questions that remain unanswered. Kenya's Political Influencer Industry Is Becoming More Sophisticated Kenyan politics already has a well-developed ecosystem of bloggers, influencers, digital strategists, anonymous accounts and political communications consultants. During major political controversies, coordinated hashtags can appear within minutes. The difference now is that artificial intelligence drastically reduces the manpower required to execute such campaigns. A communications team that previously required dozens of writers can generate enormous quantities of material using a handful of operators. Even spelling, tone and vocabulary can be deliberately varied to make posts appear to originate from people with different personalities and backgrounds. That dramatically changes the economics of political propaganda. Mass messaging becomes cheaper. Production becomes faster. And coordination becomes harder for ordinary users to recognise. 2027 Could Become Kenya's First Major AI Election The Anthropic investigation should therefore be viewed as an early warning ahead of 2027. Kenya has experienced political bots, coordinated hashtags, paid influencers, misinformation and organised online attacks during previous election cycles. The next election introduces generative artificial intelligence into that already combustible environment. Political campaigns will have access to tools capable of producing articles, tweets, Facebook posts, TikTok scripts, memes, images and videos within seconds. The technology itself is not inherently the problem. AI can legitimately help political parties communicate policies, translate messages into local languages and reach voters more efficiently. The problem begins when AI is used to disguise coordinated political communication as independent public opinion. At that point, voters are no longer simply being persuaded. They are potentially being misled about what their fellow citizens genuinely believe. Anthropic Shut Down the Operation Anthropic ultimately disrupted the Kenyan operation after identifying how its technology was being used. The company removed the accounts and organisation associated with the activity from its services and strengthened detection mechanisms intended to identify similar behaviour. There is also an important qualification: the operation did not demonstrate significant success in breaking through to genuine audiences. That should provide some reassurance. But it should not create complacency. The significance of the investigation lies in demonstrating what political operators are already experimenting with. The next network could be bigger. The next operator could be more sophisticated. And the next AI-generated campaign might be considerably harder to detect. Kenyans Must Now Question What They See Online For journalists, this development presents an especially serious challenge. A trending hashtag can no longer automatically be treated as evidence of genuine public outrage or support. Hundreds of differently worded posts do not necessarily mean hundreds of independent people reached the same conclusion. Before reporting that “Kenyans online are praising” or “Kenyans are attacking” a politician, newsrooms may increasingly need to examine whether the conversation developed naturally or was manufactured. That responsibility will become even more important as the election approaches. The Real Story Is Bigger Than Wandayi Opiyo Wandayi's appearance in Anthropic's findings makes the investigation politically significant, particularly because he is a serving Cabinet Secretary. But this story is ultimately bigger than Wandayi. There is no evidence from Anthropic's findings that he personally commissioned the operation, and that distinction should remain clear. The bigger warning concerns what has arrived in Kenyan politics. Artificial intelligence has made it possible for a relatively small number of people to manufacture political messaging at a scale that previously required armies of online operatives. One political instruction can become 50 apparently different opinions within moments. Fifty can become hundreds. Hundreds can potentially create a trending conversation. And a manufactured conversation can eventually influence a genuine one. As Kenya heads towards 2027, voters will therefore have to become considerably more sceptical about what appears on their timelines. Because the next Kenyan praising a politician or attacking an opponent online might be expressing a genuine personal opinion. Or the words might have originated from an AI system instructed to make a coordinated political campaign look like the voice of ordinary Kenyans.