Rwanda’s 2023 National AI Policy is anchored on five ethical principles - beneficence, non-maleficence, autonomy, justice and explicability. But explicability is the hardest to satisfy because it demands that those regulating an AI system understand how it works. The Policy devoted roughly 70 per cent of its substantive content to skills development. That emphasis was deliberate: build the human capacity to understand AI before writing rules to govern it. This is explicability by design: building regulatory competence in people first, so that any legislation that follows can be enforced by those with the necessary expertise. As Rwanda’s AI ecosystem grows, particularly into physical AI capable of causing bodily harm, that same deliberate approach now needs to extend from personnel to the law itself. Minister of ICT and Innovation Paula Ingabire has emphasised, both at the policy’s launch in 2023 and subsequently, that the focus on skills was intentional. Without a domestic talent base, Rwanda could not sustain an AI industry or manage the disruption it would bring. This logic extends to governance. A regulator who cannot interpret a risk assessment or understand why a model behaved in a particular way cannot meaningfully assess its compliance with principles such as non-maleficence and justice. Training engineers and data scientists, rather than only end users, builds the human capacity needed to enforce explicability as a legal principle. Where design meets danger The stakes rise sharply when AI moves beyond software into the physical world. Rwanda is attracting physical AI deployments, including data-collection partnerships linked to humanoid robotics. These activities are becoming part of the country’s AI ecosystem. A malfunctioning humanoid robot does not simply produce a flawed output; it can injure someone standing nearby. This exposure is not limited to technology developed in Rwanda. Vehicles equipped with autonomous-driving hardware are already imported and sold in the country, even though their autonomous-driving features are not certified for local use. A country does not need to develop a technology to need laws governing it; it only needs to be a jurisdiction where that technology is deployed. If such a feature were activated on a Rwandan road, questions of certification and liability would arise regardless of where the vehicle was built. Rwanda’s law currently offers no clear answer as to who would bear that responsibility. The European Union’s AI Act provides a useful comparison. Article 9 requires providers of high-risk systems to maintain a continuous risk-management process throughout a system’s lifecycle and extends equivalent obligations to deployers, not only developers. That distinction matters for Rwanda, since most of the physical AI it encounters will come from elsewhere rather than be developed domestically. At present, Rwanda has no binding equivalent for either category. Its five principles remain policy commitments rather than statutory duties. A company deploying a robot, or a dealer importing a vehicle with dormant autonomous-driving features, is therefore guided by aspiration rather than an enforceable AI-specific obligation. Rwanda has already taken an institutional step: in June 2026, Cabinet approved a National AI Agency to coordinate AI development, adoption, investment and governance across the country. As currently framed, however, its mandate appears to emphasise promotion rather than enforcement. It can convene stakeholders and guide strategy, but its authority to compel compliance is less clear. Closing that gap is the natural next step. A standalone AI Act should translate the five principles, particularly explicability, into enforceable legal standards for high-risk and physical AI. It should impose a continuous risk-assessment duty on both developers and deployers, drawing on the EU’s Article 9, and grant the agency clear implementation and enforcement powers. Rwanda built its AI governance strategy on a premise many countries overlook: a country cannot regulate a technology it does not understand, so it must first build that understanding. This approach has produced a capable workforce and an emerging institution with technical expertise. What it has not yet produced is a binding legal framework that gives that expertise legal force, particularly in addressing the physical risks posed by humanoid robots and imported vehicles with autonomous-driving capabilities, regardless of who built them. Explicability by design was the right first move. A standalone AI Act, anchored in continuous risk assessment, applicable to deployers as well as developers, and backed by effective enforcement powers, would complete that approach. The writer is President of the Rwanda Law Student Forum.
