The Evidence Gap Behind Africa’s AI Regulation Push

The Declaration Gets the Vocabulary Right

The easiest part of AI regulation is agreeing on the words everyone can endorse. Collaboration. Inclusion. Resilience. Evidence. Trust. The harder part is building the shared machinery that lets those words survive contact with markets, platforms, procurement cycles, and infrastructure bottlenecks.

That is why the signal from Lilongwe matters. The G5 summit hosted by CRASA and Malawi’s communications regulator closed with a voluntary declaration around collaborative regulation. Research ICT Africa notes that the declaration sets out six principles and seven areas of collaborative action, while anchoring itself in the ITU’s G5 framework. None of that is trivial. It means Southern African regulators are trying to treat AI, satellites, digital platforms, and next-generation networks as one joined regulatory frontier rather than separate policy files.

The weakness is also visible from the start. A voluntary declaration can coordinate attention, but it cannot by itself create market evidence, technical capacity, enforcement memory, or bargaining power. For African regulators facing platform concentration and imported infrastructure economics, the question is not whether the declaration says the right things. It is whether it changes what regulators can see together.

The Weak Point Is Not Cooperation. It Is Evidence.

The mainstream reading is generous and mostly correct: regional harmonisation is better than fragmented national rulemaking. CRASA’s own institutional mandate is to harmonise ICT regulation across the SADC region, which is exactly the kind of coordination digital markets require. Cross-border systems punish isolated oversight. A model, data broker, satellite provider, payments layer, or platform safety policy does not become local simply because a regulator’s authority stops at a border.

But harmonisation can become theatre if it standardizes language faster than it standardizes evidence. The real constraint is not a shortage of principles. It is the absence of shared measurement systems that make harm, dependency, pricing, outage risk, model deployment, data flows, and market concentration visible across jurisdictions. Without that, collaborative regulation becomes a meeting format rather than a control system.

This is the underrated part of evidence-led governance. Evidence is not just research attached to a policy memo. It is an operating layer. It determines which institutions know what, how quickly they know it, whether they can compare it, and whether private actors can be forced to answer the same question in more than one country. If every regulator has different data, different terminology, different reporting requirements, and different technical capacity, then the regulated system has the advantage before the first rule is written.

AI Turns Telecom Regulation Into Infrastructure Governance

The ITU’s policy language already points in this direction. Its development arm describes digital regulation as a set of policy and legal frameworks meant to support cross-sectoral collaboration for digital transformation. The newer phrase is “digital ecosystem builders,” used in the GSR-25 resources. That language matters because it quietly moves regulators away from the old telecom posture: license the operator, manage spectrum, monitor prices, enforce competition rules.

AI makes that posture insufficient. The relevant system now includes cloud providers, foundation model companies, app platforms, data centers, undersea cables, satellite networks, mobile operators, payment rails, government procurement, identity systems, cybersecurity standards, and labor-market intermediaries. No single regulator naturally owns that map. Yet AI deployment increasingly depends on the full map.

This is where the Lilongwe Declaration becomes more than a regional governance story. It is a test of whether communications regulators can become infrastructure interpreters. The old question was whether citizens could connect. The new question is whether institutions can understand the terms of connection: who supplies the models, where the data goes, which cloud regions host public services, what audit trails exist, how platform rules are enforced, and which firms can absorb compliance costs that local competitors cannot.

That is why this sits next to earlier Oria Veach arguments about state capacity. In state-capacity story, the point was not that optimism is wrong. It was that adoption claims mean little unless institutions can actually govern the systems they adopt. Lilongwe pushes the same problem into regional form.

Builders Should Read This as a Market Signal

For builders and investors, the temptation is to treat African AI regulation as a future compliance problem. That is too narrow. The more useful reading is that regulatory capacity is becoming part of market structure.

If evidence systems remain weak, large external platforms benefit. They already have lawyers, policy teams, telemetry, cloud relationships, and the ability to satisfy multiple regulators with polished but selective disclosure. Smaller local firms may face uncertainty without equivalent bargaining power. Governments may adopt AI tools without enough shared technical capacity to evaluate vendor claims. Civil society may be invited into consultations after the important architecture has already hardened.

If evidence systems strengthen, the market changes. Procurement can ask sharper questions. Regulators can compare incidents across borders. Competition authorities can see dependency patterns earlier. Public-sector buyers can distinguish between usable AI and vendor mythology. Local builders can compete on trust, auditability, localization, and domain expertise rather than trying to outspend global platforms.

This is also why the African Union’s continental AI strategy belongs in the same frame. Continental strategy gives the broad political horizon. Regional regulatory collaboration tests whether that horizon can become institutional muscle. The gap between the two is where most AI governance will either become real or dissolve into conference language.

The phrase “Global South voice” is often used as if presence were power. Presence matters. But the harder question, raised in the earlier piece on Global South voices, is whether participation changes the terms of decision-making. A regulator invited into a global conversation without comparable evidence infrastructure is still negotiating partly blind.

The Test Is What Regulators Can See Together

The Lilongwe Declaration should not be dismissed because it is voluntary. Voluntary coordination is often how institutional muscles form before law catches up. But it should not be overpraised either. The declaration’s value will be measured by what it makes newly observable.

Can regulators build common incident-reporting templates for AI-enabled services? Can they share market data on cloud, satellite, and platform dependencies? Can they coordinate audit expectations for vendors selling into public services? Can they track whether “localization” means local benefit or merely local sales teams? Can they compare procurement failures before each country repeats the same mistake alone?

Those questions sound procedural. They are not. They are power questions expressed as administrative design. Whoever controls the evidence layer shapes the regulatory imagination. What cannot be measured consistently becomes easy to minimize. What cannot be compared regionally becomes easy to isolate. What cannot be audited becomes a trust claim.

The undercovered story is not that Southern African regulators issued another declaration. It is that AI governance in Africa is moving into the less glamorous terrain where capacity either compounds or disappears: shared data, common definitions, enforcement memory, technical standards, and institutional trust. The declaration gets the vocabulary right. The next phase is less photogenic and more decisive.

The future of AI regulation will not be decided by who says “collaboration” most convincingly. It will be decided by who can see the system clearly enough to act before the terms are set elsewhere.