Federated learning is being sold to the world as the golden child of privacy. Keep the data where it is, only move the models, sprinkle in some secure computation, and voilà — compliance and innovation march hand in hand. In theory, it’s elegant. In practice, it is often hailed as a Privacy-Enhancing Technology, a PET that regulators and big tech alike can parade as proof of their commitment to responsible data use. But let’s ask an uncomfortable question: how realistic is federated learning as a privacy shield when viewed through the African lens? Because on this continent, the conversation isn’t just technical. It is political. It is infrastructural. It is existential.
Africa is not Silicon Valley with a tan. Here, electricity is not guaranteed. Bandwidth is expensive, and latency is not a minor inconvenience — it’s a constant chokehold on business and innovation. Federated learning assumes you can distribute model training across multiple nodes and still have reliable synchronisation, secure aggregation, and robust governance. But what happens when half your nodes are running on patchy internet that drops out every time it rains? What happens when data is held not in sleek corporate databases but in a mix of outdated servers, paper files converted into half-baked digital copies, and government registries prone to corruption or fire? Federated learning in Africa risks becoming a glamorous theory that cracks under the weight of our daily realities.
And yet, here’s the paradox: Africa needs PETs more than anywhere else. Our citizens are data-rich but power-poor. They trade their privacy for access to digital services, often without consent, often without recourse. Multinationals scoop up African data like it’s a natural resource — mined, exported, and monetised elsewhere. Regulators draft data protection laws modelled on the GDPR, but enforcement is weak and the playing field is uneven. In this environment, a PET that promises local control of data, that allows learning without exposure, could be transformative. It could redraw the power map. But only if it is implemented with eyes wide open.
Let’s be blunt: federated learning cannot be parachuted into Africa as a plug-and-play privacy fix. If we allow it to be imported wholesale from Western labs, we risk repeating the same story that has played out for decades — technologies designed for London or San Francisco that do not fit Lagos, Nairobi, or Accra. We must interrogate what “privacy” actually means in African societies, where community often outweighs the individual, and where the concept of data sovereignty is entangled with postcolonial struggles over land, minerals, and now information. Federated learning must be localised, not just translated.
Imagine a model trained on hospital data across Nigeria, Kenya, and South Africa. In theory, no patient records ever leave the hospitals; only the model updates are shared. Sounds brilliant. But who controls the central server coordinating this learning? Is it a Silicon Valley giant? Is it a government ministry with opaque oversight? Or is it an African consortium with real governance power? The answer decides whether federated learning is a tool of liberation or another digital Trojan horse.
We also cannot ignore the cost. Secure multiparty computation, differential privacy, homomorphic encryption — these are not free toys. They require computing power and skilled engineers that many African institutions do not yet have. Unless capacity is built, federated learning risks becoming the preserve of the wealthy few, widening the gap between those who can afford PETs and those who cannot. It risks entrenching digital inequality under the banner of privacy.
And here’s the final sting. Privacy is political capital. If African governments do not prioritise building their own federated learning ecosystems, others will do it for them. And when that happens, control over African data will once again be outsourced. We will be told it is for our own good, that it keeps us safe, that it protects our citizens. But protection without autonomy is just another form of dependency.
So yes, federated learning is a PET. But in Africa, it is not yet a realistic one unless it is reshaped by African hands, for African contexts, with African priorities at the centre. Otherwise, we will be left clapping for a technology that works perfectly — everywhere except here. And let’s be clear: Africa cannot afford to be the testing ground where privacy is promised but sovereignty is stolen. Either we break federated learning to rebuild it for ourselves, or we accept that once again, our data will be harvested under the guise of progress.
