Anthropic Found AI Being Used In Political Influence Operations In Africa

Anthropic, the company behind Claude, says its latest investigation has found AI being used in political influence operations targeting African audiences, including a Russian state-linked operation in the Central African Republic and a domestic campaign preparing fake grassroots political content ahead of Kenya's 2027 election.

The findings appear in Anthropic's September 2026 threat-intelligence report, which examines malicious activity involving its AI models between December 2025 and August 2026. The company says it disrupted the operations it identified and shared relevant information with authorities and industry partners.

The African cases are particularly significant because they show that AI's role in political manipulation is moving beyond simply generating text. In some cases, the technology was being used to help operate entire influence systems.

What Anthropic found in Africa

One of the most significant cases was identified in the Central African Republic. Anthropic says a Russian-speaking actor in Bangui used Claude to support a daily content operation connected to Radio Lengo Songo, an FM station broadcasting in the capital. The material was designed to promote pro-Russia and anti-France narratives while presenting the content as ordinary local journalism.

The actor also used Claude for internal tasks, including drafting employment contracts, job descriptions and systems for evaluating journalists. Anthropic says the operation was linked to Russian state-aligned influence networks and coordinated with Russian state media organisations.

This case matters because the AI-generated material was not simply being posted on an anonymous website. Anthropic says the operation used an existing radio outlet, giving the messaging access to an established audience.

The company classified the operation at Category Four on its Breakout Scale because the content was being broadcast through radio and amplified through other channels.

Kenya's 2027 election is another target

Anthropic also identified a separate operation in Kenya involving the mass production of political social-media posts. The actor used Claude to generate batches of posts designed to look like spontaneous grassroots political commentary. The content promoted government figures and attacked opposition narratives ahead of Kenya's 2027 general election.

However, there is an important distinction that Anthropic found no evidence that the campaign had successfully reached or influenced genuine audiences. It classified the operation as Category One on its Breakout Scale, meaning the activity remained confined to the artificial network Anthropic identified.

That distinction is important. The report is evidence that AI is being prepared for political manipulation, not evidence that AI has already determined the outcome of an African election.

AI is becoming an operational tool

The wider significance of Anthropic's report goes beyond individual propaganda campaigns. The company says malicious actors are increasingly using AI systems to perform multiple parts of an operation rather than simply asking a chatbot for an isolated piece of information.

Anthropic describes cases involving cyber operations, surveillance, influence operations, fraud, biological misuse, conventional weapons development and attempts to copy AI systems. The report covers Claude Haiku, Sonnet and Opus models and says the activity included suspected state-backed groups, criminals, propaganda organisations and political operators.

In other words, the concern is shifting from “Can AI generate harmful content?” to “How much of a harmful operation can AI help someone run?” That distinction could become increasingly important as AI systems become better at coordinating tasks, analysing information and producing large amounts of customised content.

Most of the operations did not go viral

There is also an important caveat that can easily disappear beneath the alarming headlines. Anthropic says many of the influence operations it discovered generated little or no authentic engagement. In several cases, the company disrupted campaigns before they had built a genuine audience.

That means the report should not be interpreted as evidence that AI has already made political disinformation unstoppable. Instead, it provides evidence of a growing production and scaling problem, generating convincing political content, fake personas and coordinated messaging can become substantially easier when AI is added to the workflow.

What does this mean for Ghana?

Ghana is not identified as a target in Anthropic's African cases. But the broader issue is relevant. Ghana launched its National Artificial Intelligence Strategy in April 2026, with the government saying the strategy will support responsible AI development and include work on ethical and legal frameworks.

The challenge is that political manipulation does not require an AI system to directly target Ghana. The same tools used elsewhere can be adopted by actors in different countries, while social-media platforms and news organisations still have to determine whether apparently authentic political content is genuine.

That makes media literacy, verification and platform-level detection increasingly important, particularly in election periods, when fabricated claims can spread faster than traditional fact-checking can respond.

The United States is beginning to debate tougher rules

Anthropic's report arrives as US lawmakers are separately considering how much responsibility should legally fall on companies developing the most capable AI systems.

Reuters reports that a group of US senators is discussing a possible legal “duty of care” for developers of advanced AI systems. The proposal remains under negotiation and there is no public bill text establishing such a requirement.

That means the US has not yet enacted the proposed framework, and it would be premature to describe the measure as an established law.

The debate nevertheless reflects a broader question that is becoming harder to avoid that when increasingly capable AI systems can be used for cyberattacks, surveillance or large-scale manipulation, how much responsibility should rest with the people building those systems?

The bigger lesson for Africa

The African cases in Anthropic's report are significant not because they prove that AI has already transformed African elections or politics. They show something more subtle. AI is lowering the cost of producing, translating, disguising and scaling political messaging, and African information environments are already part of that experiment.

The Central African Republic case shows how AI-generated material can be inserted into an existing media ecosystem. The Kenyan case shows how quickly political actors can manufacture the appearance of grassroots support, even when an operation never manages to reach a real audience.

For Ghana and other African countries, the immediate question may therefore be less about whether AI will eventually affect elections. It is whether institutions, journalists, platforms and ordinary users will be able to distinguish authentic public opinion from content manufactured to look like it.