Borrowed Ground: Kenya’s Finance Bill Chatbot and the Limits of Digital Citizenship

In June 2024, Kenyans rose against the Finance Bill in a wave of citizen-led protests. The bill was a revenue measure meant to partly fund the KES 3.9 trillion ($30.5 billion) 2024/25 national budget by raising KES 344 billion ($2.7 billion) through new tax measures. Its most controversial clauses initially targeted everyday products including a tax on bread, higher excise on mobile money and bank transfers, and a levy that would have made sanitary products more expensive. The fury began online. Within three weeks, approximately 25 million posts carrying #RejectFinanceBill2024 and its sister hashtags had moved through X (formerly Twitter). On 25 June, as parliament passed the Bill, protesters breached its precincts and the police fired live ammunition killing at least 13 people. President Ruto withdrew the Bill but the protests did not stop. Neither did state violence. The Kenya National Commission on Human Rights reported at least 60 citizens killed between June and July 2024 alongside a wave of abductions that prominently targeted visible online voices.

The withdrawal of the Bill is remembered as a vindication of digital-led protests, and in part it was. However, the digital infrastructure that aided the mobilisation did not belong to the citizens using it. My MSc dissertation followed the lifecycle of one such tool, a chatbot running on OpenAI’s servers under OpenAI’s terms, to show that dependence.

A chatbot for a tax bill

The Finance Bill 2024 was filled with legal jargon and it amended other laws by cross-reference. Tracing one clause meant consulting several other Acts, and it meant that many Kenyans either could not or did not read and understand it. In the second week of June 2024, a 24-year-old software engineer in Nairobi, defeated by the text of the bill, uploaded it into OpenAI’s custom GPT platform together with the Constitution of Kenya and the statutes the bill amended. He instructed the model to answer only from those documents and shared the result on X as the Finance Bill GPT. The post went viral and he claims the tool had roughly 150,000 users in three weeks. By rendering public information legible, the Finance Bill GPT joined a Kenyan civic-technology tradition including Mzalendo, the parliamentary monitoring site built in 2005, and Ushahidi, the crisis-mapping platform born amid the 2007-08 post-election violence. It also seeded a genre, inspiring other Generative Artificial Intelligence (GAI) tools such as Corrupt Politicians GPT to hold members of the political class accountable.

A public already in motion

Some academic accounts claimed the Finance Bill GPT and other AI tools “catalysed” the protests. The empirical evidence contradicts this claim. I interviewed thirteen Kenyans sampled to fill structurally distinct positions across the social typology of the movement and I qualitatively analysed a dataset of over two million protest-related tweets posted between June and July 2024. Tweets related to AI in any form amounted to 0.03% of the dataset and those naming a specific GAI tool just 0.01%. It is worth mentioning that these figures measure public discourse about the tools, not work done with them and certainly not the private queries to the Finance Bill GPT. Similarly, a survey of mostly Nairobi-based Gen Z respondents, analysed through structural equation modelling, found no direct path from ChatGPT use to protest participation. GAI therefore did not spark or catalyse the protests; rather it entered the protest arena as a small part of the wider digital infrastructure aiming to serve a public already in motion.

The bill and the chatbot’s replies were in English, on platforms that presuppose a smartphone and internet connection, which in Kenya means the English-literate connected urban minority. Crossing that boundary took human labour. One influential content creator I interviewed marshalled volunteers who translated and spread AI-assisted breakdowns of the Bill into 18 languages (including sign language), each checked by native speakers, economic experts and lawyers. The professionals also audited the Finance Bill GPT and judged it broadly reliable but one frontline activist remembered the chatbot informing her that Kenyans would be “taxed for having cancer”, an inaccurate compression of a real clause on hospital supplies. The “small small inaccuracies”, she told me, “were essential in building up the anger.” Scholars have shown how a loosened claim can do mobilising work when it frames a grievance in terms an aggrieved public already believes and moments of heightened contestation provide a ready audience for content trading on fear and emotion. What the AI chatbot added was “the machine heuristic”, by which machines are presumed more accurate and objective than people, meaning that an interpretation issued by an AI model entered the conversation pre-credentialed.

The takedown

Late in June 2024, the software engineer got a call from a withheld number warning him to consider the consequences of what he was doing. Such a warning works by leaving the caller, the consequence and the authority behind it unnamed. It can afford to. Achille Mbembe describes the postcolonial commandement, a power that stages itself so insistently that its subjects learn to read it without being told. A threat addressed to such a subject arrives already understood. “I actually got scared,” the software engineer told me and he started feeling like he was being followed. Soon afterwards, OpenAI removed the Finance Bill GPT from public access citing an unspecified breach of its terms. When he tried to appeal, he received an email from OpenAI repeating that the tool had been found to violate the company’s “Terms of Use or Usage Policies”. It named no term and offered no evidence noting only that the tool would continue to be available to the creator to chat with privately.

The software engineer believes the Kenyan state procured the takedown reasoning that no one else wanted the conversation quelled. This cannot be empirically proven and there could be other plausible alternative explanations including automated moderation triggered by content near the political-process categories OpenAI had restricted for the US election year, its policies barring applications for “political campaigning and lobbying”, user reports in volume through the reporting channel OpenAI had built into GPTs in January 2024, or an internal trust-and-safety decision. Regardless, all these alternatives produce the same observable: a tool that vanishes from the public and an appeal answered with boilerplate. In each of these scenarios, the actor that exercised the power of removal was the AI provider. If the Kenyan state was the prime mover as the software engineer claims, it acted by borrowing the provider’s authority. 

Tenure

Digital citizenship is essentially organised around two questions: access, whether people can reach the digital sphere and exercise citizenship online; and acts, what they do and what rights they claim once there. Researching the 2024 protests in Kenya, I became more interested in a third dimension: tenure, the security of a citizen’s hold on the digital infrastructure through which civic acts are performed. The “sovereigns of cyberspace” who control the digital space can and do alter its affordances unilaterally as was the case with the Finance Bill GPT. Which terms or clauses were enforced to remove it were not revealed, an opacity that can result in control over political speech exercised without democratic accountability. Francis Nyamnjoh warned that platform citizenship “can be withdrawn at the whim and caprice of the platform provider”, and in this case it was. The deleted Finance Bill GPT revealed the condition under which digital citizenship is mostly conducted, on borrowed ground, held until the owner wants it back. What would change that condition – collective or public provisioning of digital infrastructure and platform capacity, and strong regulation able to guardrail powerful tech companies – lies beyond any single movement.

Mohamed Amin Abdishukri Maalim completed the MSc in African Studies at the University of Oxford in 2026 with distinction. This post draws partially on his MSc dissertation titled “Generative Artificial Intelligence, Digital Activism and State Response: Tenure and the Limits of Digital Citizenship in Kenya’s Anti-Finance Bill Protests”.