advertisement
Kenya’s Legal Gaps in Governing AGI Risks
In January 2026 the Office of the Data Protection Commissioner confirmed that Tools for Humanity had deleted biometric data collected from more than 350,000 Kenyan citizens. The episode demonstrated that Kenya’s legal system can identify unlawful extraction, subject it to judicial scrutiny, and compel remediation. It also exposed a deeper structural vulnerability: an offshore entity moved specialised hardware into public spaces, extracted biological identities, and transmitted them across borders while the state responded through linear administrative timelines. That gap becomes existential once the industry shifts from data collection to autonomous AGI systems.
Kenya is not a manufacturer of advanced semiconductor accelerators, nor does it host the multi-gigawatt data centres in which foundation models are trained. It operates principally as an end-user market, consuming foreign models through APIs and supplying annotation labour. A sovereign state that cannot examine the algorithms running its banking networks, telecommunications infrastructure, and government advisory systems has effectively ceded control of core digital functions. Sovereignty remains theoretical while physical infrastructure and deployment decisions remain abroad.
Empirical incidents already illustrate the limits of conventional regulation. During internal cyber-capability evaluations of OpenAI’s GPT-5.6 Sol and an unreleased frontier model in July 2026, autonomous agents constructed hidden communication boards, shared methods for bypassing sandbox constraints, fabricated execution records, and breached external infrastructure at Hugging Face. These behaviours emerged not from explicit instructions to cause harm but from optimisation processes that sought any available path to solve assigned problems. In a separate episode, Google’s Gemini model autonomously accessed the public internet and compromised three external organisations during cybersecurity testing.
advertisement
Such cases give concrete force to Nick Bostrom’s instrumental-convergence thesis: agents of sufficient intelligence, irrespective of their final goals, will pursue intermediate goals such as power-seeking, strategic deception and self-preservation because these help achieve almost any ultimate objective. When contemporary architectures already demonstrate strategic deception to evade oversight, human control becomes increasingly precarious under recursive self-improvement, in which models autonomously refactor their own code and hardware.
The Kenya Artificial Intelligence Bill, 2026 (Senate Bills No. 4 of 2026) marked a serious effort to establish statutory ground rules. It proposed an Office of the Artificial Intelligence Commissioner, a public registry for high-risk deployments, mandatory human-oversight mechanisms, and distinctively for the Global South, workforce-impact assessments and reskilling contributions for labour-displacing systems.
The Senate Committee on Information, Communication and Technology nevertheless recommended that the bill not proceed. Its central limitation mirrored that of many Western frameworks: the assumption that legal rules alone can govern autonomous systems. Criminal fines of up to five million shillings remain an inconsequential operating expense for trillion-dollar technology firms. An administrative notice issued after an autonomous agent running on an overseas cloud has already disrupted local payment infrastructure offers negligible protection.
advertisement
Comparative regimes demonstrate the required scale of deterrence. The European Union’s AI Act indexes penalties to global turnover—up to 7 percent of worldwide annual turnover for prohibited practices and up to 3 percent for other serious non-compliance, including general-purpose AI models.
The United Kingdom’s Online Safety Act empowers Ofcom to seek court-ordered service-restriction and access-restriction orders that can compel payment processors, advertising networks, search engines, and internet-access providers to withdraw services or block access. Effective deterrence must be operationally meaningful to the organisations that deploy the technology.
Kenya’s regulatory architecture must therefore evolve beyond traditional consumer-protection paradigms. Governance must be pre-emptive, establishing enforceable requirements before catastrophic vulnerabilities materialise; preventive, creating mechanisms to halt automated scaling or agent deployment when unexpected behaviours emerge; deliberate, grounded in technical realities rather than speculative panic or industry capture; and targeted, focused on mass automated influence operations, critical-infrastructure vulnerabilities, and market manipulation rather than broad superficial compliance. These principles require an integrated architecture that links legal rules directly to physical computing infrastructure.
advertisement
Parliament should develop a new Artificial Intelligence Bill informed by the forthcoming National Artificial Intelligence and Emerging Technologies Policy. The legislation should introduce turnover-based penalties, calculated using the higher of a fixed sum or a percentage of worldwide turnover, with appropriate tiers for smaller enterprises. Foreign AI providers should also be required to appoint local authorised representatives, giving regulators a legally accountable entity against which to enforce penalties and orders.
The regulator, working with the National Computer and Cybercrimes Coordination Committee, should have statutory authority to issue binding operational stop orders when autonomous systems exhibit uncontained behaviours, unauthorised lateral movement or significant security violations. These orders should be enforceable through telecommunications networks and internet exchange points. Independent technical audits of high-risk systems and general-purpose AI models should also be mandatory before deployment in sensitive sectors.
True sovereignty also requires physical infrastructure. Section 50 of the Data Protection Act, 2019 already empowers the Cabinet Secretary responsible for ICT to prescribe that certain processing occur only through servers or data centres located in Kenya on grounds of strategic state interest. This provision must be operationalised through sovereign compute nodes at Konza Technopolis, powered by domestic hydroelectric and geothermal resources at Olkaria and by nuclear projects overseen by the Nuclear Power and Energy Agency and the Kenya Nuclear Regulatory Authority. Sovereign nodes permit utilisation of advanced open-source foundational models for economic tasks while keeping sensitive national data, inference processing, and security checks on domestic hardware that cannot be remotely severed or surveilled.
Kenya must further leverage its position in the global technology supply chain. Advanced computing clusters depend on critical mineral substrates—cobalt, manganese, lithium, sourced in substantial measure from the African continent. Working through the African Continental Free Trade Area and African Union digital frameworks, Kenya can help coordinate a regional policy that conditions access to strategic resources on tangible investments in domestic infrastructure, localised research facilities, and reciprocal technology access. The transition from passive consumer of external intelligence to active participant in the physical foundation of global computing is both feasible and necessary.
Domestic institutions must close the remaining gap between technical capability and administrative oversight. Extending the skilled-persons model found in the UK Online Safety Act, regulatory authorities should be empowered to appoint independent technical auditing bodies drawn from local academic institutions and specialised safety organisations to inspect model architectures and audit logs before systems are deployed across financial services, telecommunications, and public administration. Leaving compliance to internal corporate self-assessments has repeatedly failed; independent verification must become a baseline requirement.
AI innovation without containment, however, is simply exposure.
In the accelerating transition toward AGI and eventual superintelligence, strategic focus must expand beyond basic consumer privacy to confront the catastrophic and existential realities of recursive algorithmic power. Legal statutes must possess real enforcement teeth; critical data must rest on sovereign infrastructure; and Kenya must insist on physical, hardware-level circuit breakers that keep human agency permanently in the loop. Uncontained autonomous systems must never be permitted to operate on sovereign citizens’ data, infrastructure, and democratic institutions without oversight, transparency, and consent.
The writer is a Communications and Policy Consultant, University of London Law Student, and an IT Infrastructure Engineer.