Tech Explained: Anthropic is Becoming the Backbone of Rwanda’s Government. But Who is Accountable? in Simple Terms

Anthropic in February signed a three-year memorandum of understanding with the government of Rwanda to embed its artificial intelligence systems across the country’s health ministry, public sector agencies and education system.

The agreement reveals a troubling reality, as no external review mechanism — whether a parliamentary review body, a multilateral oversight process or a civil-society disclosure requirement — was triggered, because none exists for commercial AI partnership of this kind. No framework, anywhere, required one. The question that absence raises does not end in Kigali.

The MOU covers three areas: support for the health ministry’s campaign to eliminate cervical cancer and reduce malaria deaths, Claude and Claude Code access for government developers and the expansion of an education initiative that puts AI learning tools in the hands of students across eight African countries. The MOU is non-binding. The infrastructure it sets in motion is not.

The agreement formalizes and extends a partnership Anthropic and tech training service ALX announced in November to deploy Chidi — a learning companion built on Claude — to hundreds of thousands of students across Africa.

This all means that government developers are now being trained on Claude Code with API credits provided by Anthropic. Health ministry workflows are being designed around Anthropic’s model. Each of those decisions, made individually, may seem reasonable. Taken together, they describe a situation where a private company has become load-bearing infrastructure across three of the most sensitive domains of Rwandan public life — before any governance framework has defined what obligations Anthropic should face.

If the terms of the deal are non-binding but the dependency is not, who has standing to renegotiate? Who is accountable if the model is deprecated, if pricing for the products shift or if data-handling practices change?

Anyone who watched AWS and Azure become the default backbone of government IT across the developing world in the 2010s — without any meaningful accountability discussion about what that meant — will recognize what comes next.

This is not a criticism of Rwanda. Kigali is one of the most deliberate governments in East Africa when it comes to technology adoption. Its $200 million Digital Acceleration Project, co-funded by the World Bank and the Asian Infrastructure Investment Bank, is already more than halfway complete. Rwanda set a target to eliminate cervical cancer by 2027, three years ahead of the WHO’s global timeline.

Rwanda is not stumbling into AI dependency. It is making a calculated bet — attract major technology partners early, position as a regional first mover, draw further investment. The question is not whether that bet is reasonable. It is whether anyone outside Rwanda and Anthropic gets to weigh in on its terms.

Rwanda will likely see real benefits from this partnership — better decision-support tools for health workers, AI access for students who would otherwise have none and developers building on infrastructure that didn’t exist before.

Elizabeth Kelly, the company’s head of Beneficial Deployments, emphasized training and local autonomy, and Anthropic has public commitments on AI safety that go further than most. But responsible corporate behavior is not a substitute for governance. A company can be entirely well-intentioned and still create an accountability vacuum by operating in a space where no external oversight exists — which is why good intentions cannot close a structural gap.

Minister of ICT Paula Ingabire called the agreement “an important milestone in Rwanda’s AI journey.” But milestones are easier to evaluate when someone other than the parties to the deal gets to read the map. Nothing required that before these terms were signed. Civil society had no standing to review the arrangement. The agreement contains no disclosure provisions specifying what happens to Rwanda’s medical-surveillance priorities if Anthropic’s commercial strategy changes.

The pattern has precedent. Huawei built roughly 70 percent of Africa’s 4G network through commercial deals between a private company and African governments — and those dependencies are still being contested by governments that now find it politically and financially prohibitive to switch.

For years, Western governments criticized African states for those choices while building no credible governance alternative that might have guided different decisions. The lesson is not about which country’s company is doing the embedding. It is about what happens when commercial infrastructure deployment outpaces governance: the dependency becomes politically irreversible before anyone has decided whether it should exist.

This is not a problem unique to AI, but the absence of governance here is more striking because comparable sectors have already solved it.

When pharmaceutical companies run clinical trials in partnership with government health ministries, institutional review boards exist to protect the interests of people who are not party to the commercial agreement. When foreign companies access government financial systems, disclosure and audit requirements apply regardless of whether the partnership is well-intentioned.

The argument that an AI company embedding itself in a national health infrastructure needs no equivalent mechanism is not a neutral position. It is a policy choice — made by default by everyone who has not yet decided it needs to be made deliberately.

If the next deal involves a government with less technical capacity to evaluate what it is agreeing to, or an AI system with fewer safety commitments, the same accountability vacuum applies — with higher stakes and less recourse.

What happens when the partnership is not in health and education but in law enforcement or surveillance? Across the continent, commercial AI partnerships with governments have already moved in that direction, spawning surveillance systems that are sold as public-safety tools and repurposed for tracking political opposition.

The African Union endorsed a Continental AI Strategy in July 2024, and in April the Global AI Summit on Africa in Kigali produced an Africa Declaration on AI,which committed all 55 African Union member states to principles of AI sovereignty, data governance and responsible deployment — and proposed a $60 billion Africa AI Fund.

What the strategy has not yet done is apply its governance ambitions to the commercial partnerships already reshaping African public infrastructure. That gap is worth naming precisely: the MOU is non-binding because that is how commercial tech partnerships work. No one at Anthropic or in Rwanda’s ICT ministry did anything unusual by signing a non-binding agreement.

The accountability vacuum is not a failure of this deal — it is the normal condition of how private AI companies engage with governments everywhere.

Disclosure requirements when companies embed in government health or education systems, standard terms around model deprecation and pricing continuity, some form of external review before public institutions build core workflows on commercial AI — none of that requires a new treaty. It requires the African Union’s existing governance work, declarations written and frameworks endorsed, to treat private-sector partnerships as the main problem, not an afterthought.

With the frameworks in place and deals being signed, whether governance arrives before the dependencies harden is the question the strategy now exists to answer.