By Brian Byaruhanga |
Kenya was reported, in a recent global update, to hold the world’s highest rate of Artificial Intelligence (AI) tools usage – 42.1% of surveyed internet-using adults. The figure has travelled quickly. It now adds to a familiar story: Kenya as Africa’s AI frontrunner, sprinting ahead of its regulatory infrastructure, in need of a comprehensive AI law to close the gap. While the claim might seem true, is the story built around it accurate?
The 42.1% does not measure Kenyan AI capability. It measures Kenyan consumption of AI built elsewhere. The chatbots – OpenAI’s GPT, Google’s Gemini, and Anthropic’s Claude – run on compute owned by foreign firms, are trained on data scraped without Kenyan consent, and are monetised in jurisdictions outside Kenya. To call this “adoption” is to mistake dependence for agency. The frontrunner framing flatters us into thinking Kenya is racing, yet, in reality, it is being driven.
Furthermore, AI-powered tools in current Kenyan usage are not only chatbots. They include algorithmic feeds – TikTok’s For You Page, Instagram’s Reels, X’s recommended timeline, YouTube’s autoplay, and Facebook’s News Feed, used by the country’s 18.4 million social media users. These algorithms shape the perception of information consumed by every Kenyan on social media. Most users do not experience or recognise these systems as AI. They experience them as the internet. That is the most pervasive form of AI adoption: Kenyans do not need to log into an AI chatbot to be governed by foreign machine-learning systems; they need only to scroll.
In 2022, a Mozilla investigation by Odanga Madung showed how TikTok’s For You Page algorithm boosted election disinformation in Kenya far beyond the reach of any individual post – proof that recommender systems are themselves political infrastructure. The 2025 State of Internet Freedom in Africa report flags this issue, directly pointing out that algorithmic recommendations, content curation, and automated moderation “profoundly affect how citizens access news, engage politically, and mobilise digitally.” Yet, this usage is not factored in as part of the 42.1%, and neither does the recently proposed Kenya AI Bill 2026 provide an oversight or regulatory mechanism to address the potential harms.
This is where Kenya’s emerging AI governance architecture, as outlined in the Bill, the National AI Strategy, and the regulation-versus-innovation commentary that frames them, falls short. In treating AI as a commodity, Kenya is integrating – a thing to be permitted, audited, and made fair. But the labour that trains these systems is Kenyan: the data annotators in Nairobi’s outsourcing centres, the moderators who absorbed the worst of OpenAI’s training data for less than two dollars an hour, and the gig workers whose human feedback shapes what foreign models call “alignment”. That extraction predates the 42.1% figure and, in many ways, produces it, yet safeguarding the risks to these workers has not been prioritised.
In June 2024, during the #RejectFinanceBill protests, Kenya experienced a nationwide internet disruption that the Communications Authority denied it planned, which telecoms attributed to undersea cable cuts, and that NetBlocks and Cloudflare confirmed. The Kenya National Commission on Human Rights subsequently documented at least 82 abductions and enforced disappearances of digital organisers. These events were routinely framed as exceptions, an unfortunate moment of overreach.
The internet disruptions and the abductions were not just exceptions. They were fundamental to the design of Kenya’s executive-security-telecoms nexus, i.e., the Interior and ICT ministries, the Communications Authority, the National Intelligence Service, the National Cyber Crimes Coordination Committee (NC4), the Directorate of Criminal Investigations (DCI), and some licensed mobile operators and internet service providers (ISPs) acting in concert as a single discretionary instrument of the state. An apparatus that can track and disappear activists and protestors, deny a shutdown that it instituted, is the same apparatus that will play a key role in determining what AI deployments are permitted and how citizens can use it safely.
The same Communications Authority that suspended Telegram sits inside the country’s AI governance ecosystem. The same security apparatus that disappeared activists is responsible for the biometric surveillance systems that the National AI Strategy declines to prohibit. These institutions will oversee the enforcement of any AI law passed in 2026. A regulatory framework that does not address the behaviour of its enforcers is flawed.
Another shortfall is the plea in nearly all Kenyan AI discourse: to “balance innovation and regulation”. Innovation in Kenya is not endangered by regulation. It is endangered by foreign capital concentration, undersea cable bottlenecks, and the migration of local talent to international firms. Far from being a hindrance, regulation is a vital tool for oversight and asserting sovereignty. Instead of weighing regulation against innovation, the focus should be on the tension between sovereignty and access – two areas currently dominated by the same powerful interests and left unprotected by frameworks that fail to identify extraction for what it truly is.
What, then, does honest governance look like?
AI Governance for Kenya and Africa as a whole should look like a gate, a pre-deployment review mechanism where data access is a privilege earned through evidence rather than a courtesy extended in advance. The non-negotiable is an independent authority with the power to halt or redesign deployment when non-compliance is found. Not advised. This should apply as much to recommender systems already operating inside Kenya as to new models entering the market. An AI law that cannot reach the algorithmic curation layer of TikTok, Meta, and X has already exempted the largest category of AI affecting Kenyans. The current draft of the AI Bill fails to establish such authority.
While apprehensions regarding censorship, over-regulation and enforcement capacity are valid, the lack of a halt mechanism remains a more critical flaw. Although the draft National AI Strategy mentions governance, it fails to identify the ultimate decision-maker. Until that authority is clearly defined, Kenyans are merely debating form rather than addressing the core substance.
Kenya does not have an AI regulation gap. It has an accountability gap, and AI is the new vector through which that gap widens. The proper question is not ‘How do we govern AI?’ but ‘How do we govern the institutions that will govern AI?’ A “human-centred design” is not a complete answer. The answer begins by rejecting the frontrunner story, naming the extraction, tying the June 2024 shutdown to the architecture rather than the cable, and reserving the word ‘sovereignty’ for governance that can actually halt an AI deployment when the evidence requires it. Ultimately, introducing new AI laws into a system that already lacks institutional checks and balances will not protect citizens; rather, AI will just become a new vector through which state overreach expands unchecked.