KENYA FLAGGED OVER AI MISUSE IN POLITICAL PROPAGANDA

Kenya has been named in a new global threat intelligence report by artificial intelligence company Anthropic after investigators uncovered a domestic operation using its Claude AI system to produce fake political content ahead of the 2027 General Election.

The operation, identified as GTG-54004, involved a single actor generating batches of 50 social media posts at a time and presenting them as ordinary grassroots opinions. Anthropic said the account and organisation behind the activity were subsequently removed.

The findings are contained in Anthropic’s “Detecting and Countering Misuse of AI: September 2026” report, published on September 10 and covering malicious activities detected and disrupted between December 2025 and August 2026.

According to the report, the operator supplied Claude with political topics, talking points and hashtags before asking it to turn them into posts that appeared spontaneous rather than coordinated.

A substantial portion of the content promoted Energy and Petroleum Cabinet Secretary Opiyo Wandayi, particularly following his role in halting a planned Kenya Power electricity tariff increase. The posts used hashtags including #PowerReliefKE and #PoweringTheNewKenya.

The tariff issue had already attracted public attention after Kenya Power applied on March 31, 2026, to increase retail electricity rates. Wandayi later announced on June 3 that the application had been withdrawn, saying existing retail tariffs would remain unchanged unless lawfully reviewed under the Energy Act.

The AI-generated campaign also sought to influence perceptions of the opposition. Anthropic said the network circulated claims that Kenya’s United Opposition coalition was falling apart ahead of the 2027 election, specifically mentioning former Deputy President Rigathi Gachagua and former President Uhuru Kenyatta.

Anthropic assessed the political messaging as broadly pro-administration and said the hashtags and tone were plausibly aligned with the ruling coalition. However, the company stressed that it found no evidence of government involvement and did not establish the identity of the organisation responsible.

Instead, it described the activity as a domestic Kenyan political astroturfing operation.

The case also had a commercial side. Anthropic said the same AI workflow was used to market Kenyan retail brands through the online personas “SHANKI” and “Elkins Marketer.”

In one example, a promotional message was converted into posts made to look like organic social media commentary, with links inserted into every fifth post.

The discovery illustrates how the same tools used for ordinary marketing can be adapted to political influence campaigns. Anthropic said Claude's principal contribution was not developing the political strategy, but allowing the operator to produce large volumes of content while making it appear to come from different ordinary users.

The underlying political messages had already been determined by the people behind the operation before Claude was used.

Despite the scale of the production process, Anthropic gave the Kenyan operation a Category One rating on its Breakout Scale, the lowest level in its framework. The company said the activity remained within a network of fake accounts and did not reach or influence genuine users on the platform.

The operation was eventually detected following a tip from OpenAI concerning recurring activity on its own platform. Anthropic said the information prompted an internal investigation into suspected coordinated inauthentic behaviour in Kenya, leading to the removal of the account and organisation and the development of new detection measures based on the campaign's behaviour.

The incident also highlights a broader problem that predates generative AI in Kenya: organised attempts to manufacture online public opinion.

Research published by the Mozilla Foundation in 2021 documented coordinated Twitter campaigns involving thousands of accounts and tens of thousands of posts targeting Kenyan journalists, judges and civil society actors. The research also found cases where people were paid to push particular hashtags and narratives.

What has changed with generative AI is the speed and scale at which similar content can be produced. Instead of relying entirely on large groups of people to write individual posts, an operator can use AI to prepare large batches of material that are then distributed through social media accounts.

Anthropic said its Kenyan case forms part of a wider chapter covering nine influence operations linked to actors in different parts of the world, including Russia, Iran, Turkey, the Gulf, South Asia, Africa and Europe.

In another African case, the company described a Russian-linked propaganda operation involving a radio station in Bangui, Central African Republic, which it associated with the Wagner-linked Africa Corps network. It also documented a separate commercial influence operation that produced thousands of politically slanted articles across about 70 fake news websites, including content concerning the Democratic Republic of Congo's dispute with Rwanda.

Anthropic noted that its visibility into such operations generally ends once AI-generated material leaves its systems and appears on social media. However, the company said monitoring activity at the content-production stage can provide an opportunity to disrupt campaigns before they gain a genuine audience.

The Kenyan case comes as the country works on a broader framework for regulating artificial intelligence. Kenya's proposed Artificial Intelligence Bill 2026, sponsored by Senator Karen Nyamu, seeks to establish an Office of the AI Commissioner and introduce rules governing different categories of AI systems.

The proposed framework would classify AI applications into four broad risk levels: unacceptable, high-risk, limited-risk and minimal-risk systems. High-risk applications, including those used in areas such as healthcare, education, banking, biometrics and elections, would face greater regulatory scrutiny.

The proposed law also addresses the use of people's voices, images and likenesses, including requirements around consent, while seeking to provide controlled environments known as regulatory sandboxes where technology developers can test AI systems under supervision.

Another proposed measure would require organisations introducing AI with significant potential to affect employment to assess its impact on workers and consider measures such as reskilling.

The legislation has nevertheless generated debate. Nyamu has argued for tougher penalties, while technology stakeholders have raised concerns that excessive regulation and compliance costs could place a heavier burden on smaller Kenyan technology companies.

The latest Anthropic findings therefore arrive at a sensitive time for Kenya, with the 2027 election approaching and AI becoming increasingly embedded in political communication, marketing and online media.

For voters, the episode is also a reminder that a busy social media timeline does not necessarily represent genuine public sentiment. Repeated messages, identical hashtags, unusually coordinated posting and accounts with little evidence of ordinary activity can all warrant closer scrutiny.

Anthropic's investigation did not establish that the campaign succeeded in changing public opinion. But it demonstrated how artificial intelligence can make an organised messaging operation appear much larger and more spontaneous than it actually is — raising new questions about the reliability of online political conversations as Kenya moves toward the 2027 polls.