Rwanda Sets Up National AI Agency in Bid to Become Africa's AI Hub

Rwanda Sets Up National AI Agency in Bid to Become Africa's AI Hub

Rwanda's Cabinet approved a National Artificial Intelligence Agency on June 8, 2026. On October 27, 2026, President Paul Kagame appointed its first senior officials: Olivier Twagirayezu as Chief Operations Officer, Jessica Ingabire as Chief Delivery Officer and Robert Kayinamura as Chief Strategy and Partnerships Officer. For banks, fintechs, telecoms and foreign investors, the question is now who will govern AI in Rwanda, under which rules, and how soon. An agency, not yet a regulator Many reports call the Agency a regulator. So far, it is not one. Its reported mandate covers research, skills, AI ethics and safety, industry development and international partnerships. It takes over the work of the AI Office in the Ministry of ICT and Innovation and builds on the Rwanda AI Scaling Hub, which has raised about RWF 25 billion, close to US$17 million. Nothing reported so far lets the Agency license AI systems, inspect companies or impose fines, and the instrument setting out its powers has not been reported. Until it is, the Agency's standards will not bind anyone directly. They will still count, because public tenders, regulators and investors will use them as the benchmark. Companies that sell to government will feel this first. The regulators who already have teeth The real exposure today lies with the National Cyber Security Authority. Law No. 058/2021 of October 13, 2021 relating to the protection of personal data and privacy (the "Data Protection Act") has applied in full since its two-year compliance period ended on October 15, 2023, and it covers any AI system that handles personal data. Four of its rules matter most. Every data controller and processor must register, and the certificate takes up to 30 working days to issue. Personal data must stay in Rwanda unless the certificate allows storage abroad, and a transfer abroad needs a legal ground and a written contract. Since most AI tools run on foreign servers, entering customer data into one is a cross-border transfer. A person may refuse a decision made solely by a machine where it seriously affects them, which covers automated credit scoring, hiring and insurance pricing. And a data breach must be reported within 48 hours, even when it happens at a vendor. The penalties are heavy. A company operating without a certificate faces a fine of 1% of its global turnover for the previous year (Article 53). A company convicted of an offence faces 5% of annual turnover (Article 62). A foreign provider that handles the data of people in Rwanda must appoint a representative in the country (Article 39). Financial firms face a second law. Law No. 023/2026 of May 25, 2026 regulating virtual asset business (the "Law on Virtual Asset Business") allows only licensed companies to operate, under the Capital Market Authority, the National Bank of Rwanda and the Financial Intelligence Centre. Licence holders must verify customers, secure their systems and meet reporting duties. If they use AI for these tasks, they answer for its mistakes. Why Rwanda's model is different The European Union wrote an AI law first. The United States has left AI to its states and sector regulators. Rwanda has built the institution before the statute, and kept the binding rules in laws it already has. This gives investors one office to deal with on AI, as the Rwanda Development Board does for company registration. That office does not yet enforce anything, so an investor who speaks only to the Agency has covered the opportunity and missed the risk. It also means the standards are still unwritten, and companies here can help shape them. What companies should do now Start with a list. Record every AI tool the company uses, including chatbots staff use on their own, what personal data goes into each, and where it is hosted. Then check that list against the registration certificate. If the certificate does not cover the processing or the storage abroad, apply to amend it before the tool goes live. Next, look at any system that makes decisions about people. Add a human review and a way for customers to contest the result, and complete a data protection impact assessment before launch. Then fix the vendor contracts. They should say whether the vendor may train on your data, who pays for a wrong output, and how fast the vendor must report a breach. A contract that gives the vendor 30 days leaves the company unable to meet its own 48-hour deadline. An example shows the cost. Suppose a Kigali lender buys a foreign-hosted credit-scoring tool that approves loans with no loan officer involved. Two weeks before launch, a legal review finds that the certificate does not allow storage abroad, no impact assessment exists, and declined applicants cannot ask for a human review. Amending the certificate alone can take 30 working days, so the launch slips by more than a month. The questions still open Three things are not yet known: whether the Agency will get enforcement powers, whether Rwanda will pass a dedicated AI law, and how the Agency will share its role with the National Cyber Security Authority and the sector regulators. The publication of the Agency's founding instrument and of its first standards will answer them. Until then, the companies best placed to benefit from Rwanda's AI plans are those that already comply with the law in force.