This month, Kenyans learnt from technology company Anthropic that unidentified actors had been using artificial intelligence to create the appearance of grassroots political support ahead of the 2027 election. In its September 2026 misuse report, Anthropic says a pro-government operator used its Claude models to prepare fake grassroots social media posts ahead of the country’s 2027 general election. The actor generated batches of exactly 50 tweets at a time and explicitly instructed the model to make the posts look like organic commentary and not a coordinated campaign.
The content was predictable. Many posts praised Energy Cabinet Secretary Opiyo Wandayi for halting a scheduled Kenya Power tariff increase. Anthropic did not, however, identify evidence linking Wandayi or his ministry to the operation. Wandayi has publicly denied that he or the ministry authorized or participated in the campaign.
The company rated the operation at the lowest level of its impact scale because the activity stayed inside the network of fake accounts and did not reach or influence real people. That is reassuring about this operation but tells us little about the fact that we only know about this case largely by chance. Anthropic says it investigated after OpenAI shared a tip about recidivist activity on its platform, and the company acknowledges its visibility typically ends once content leaves its platform. Voluntary disclosures by private firms are welcome, but they do not amount to an election safeguard.
This matters more now because Kenya has just concluded an AI cooperation agreement with Anthropic, with Foreign Affairs Principal Secretary Korir Sing’Oei among the officials involved. Kenyans deserve to know whether that agreement says anything about election integrity, and whether similar commitments will be sought from every AI company operating here. If the agreement includes commitments relevant to the development, deployment or governance of AI in Kenya, the government should make those commitments public. Any framework for responsible AI should apply consistently and should include meaningful human-rights safeguards.
The Kenyan election body is not blind to this. IEBC’s Election Operations Plan 2025-27 names AI-driven misinformation and deepfakes as emerging risks that could undermine public confidence.The commission has said it will strengthen partnerships with platforms and media regulators to rapidly identify and remove harmful content. NCIC, for its part, leads a new High-Level Multi-Agency Steering Committee coordinating intelligence sharing, investigations and prosecutions on hate speech and electoral offenses, and plans to work with the Media Council of Kenya and the DCI cybercrime unit to monitor political rhetoric across rallies, radio and social media.
The challenge is ensuring that efforts to address coordnated and deceptive political influence do not sweep up legitimate political expression. Kenya has been here before. Kenya has already seen the dangers of responding to harmful online content through overly broad legal powers. Courts have raised concerns about provisions of the cybercrimes framework that risk restricting freedom of expression. Any new measures addressing AI-enabled political manipulation must therefore be clearly grounded in law, pursue a legitimate aim, and be necessary and proportionate, with adequate safeguards against arbitrary or selective enforcement.
Recommendations to IEBC
Mandate that campaigns disclose when they use AI. The IEBC should develop clear, rights respecting rules requiring political actors to disclose the use of materially synthetic or manipulated content where it could reasonably mislead voters about its origin or authenticity. The rules should also require clear disclosure of paid political influencer and coordinated campaign activity, so voters can distinguish political advertising and organised campaigning from independent grassroots expression.
Release the platform partnership. Any agreement that the IEBC enters into with social media or AI companies should be publicly accessible, subject to narrowly defined and lawful protections for genuinely confident information. The IEBC should regularly publish information on the operation of such agreements, including the number and nature of referrals, the grounds on which action was requested, the action taken and mechanisms for review or appeal. Calls for rapid removal of political content must be accompanied by transparency, clear rules, independent oversight and accessible mechanisms to challenge decisions. Otherwise, efforts to combat harmful content can undermine freedom of expression and access to information.
Recommendations to NCIC
First, NCIC should strengthen its capacity to assess and verify potentially manipulated audio, video and other digital content, including through partnerships with independent fact-checkers and appropriately qualified forensic experts. Any assessment should be based on transparent methodology and should distinguish questions of authenticity from questions of whether content meets the legal threshold for hate speech or another offence. Uncertainty about the provenance of digital content should not prevent authorities from investigating credible allegations, but neither should suspicion that content was AI-generated be treated as evidence that it is unlawful.
Second, NCIC’s monitoring should distinguish between legitimate political coordination and networks deliberately designed to deceive the public about the identity, independence or scale of political support. Coordinated deceptive networks can amplify ethnic or political tensions even where individual posts may not, on their own, meet the legal threshold for hate speech. Any response should therefore focus on demonstrable lawful conduct while respecting legitimate political expression and association. Where there are credible indications that an organized influence operation has been commissioned or financed, authorities should investigate those links in accordance with the law and with appropriate judicial and procedural safeguards.
The General Election is set to happen in less than a year. The operation was identified and disrupted before Anthropic found evidence that it had reached an authentic audience. The operation demonstrates how AI can reduce the cost and increase the scale of coordinated political manipulation. As these technologies develop, authorities and platforms will need to strengthen their safeguards without creating new tools for censorship or surveillance. The IEBC and the NCIC have months ahead to develop rules that will open up manufactured opinion without touching real political debate. Kenyans are entitled to both.
The human-rights concern is not simply whether political content is true or false. Elections depend on people’s ability to form opinions, access diverse information and participate freely in public affairs. Attempts to manufacture the appearance of public support can interfere with that environment, but measures adopted to address them must themselves respect freedom of expression, association, privacy and access to information.