Why Africa needs its own AI off switch

In the early hours of a weekday morning, merchants in Nairobi's wholesale markets are delivering fresh produce to retailers, while customers utilize the mobile money platform, M-Pesa, to complete their transactions. However, imagine a sudden system crash in the market, bringing economic activity to a halt. This scenario is not just hypothetical, as it could easily play out in Accra, where MTN's MoMo system also crashed, causing a significant halt in business at the Makola Market.

With the escalating number of AI-driven cyberattacks, these are plausible threats in Africa's digital landscape.

Currently, discussions around AI agents in Africa revolve around global rivalries, such as China versus the United States, and the debate between open and closed models. However, the most critical aspect for African nations is ensuring their AI systems can be controlled locally. The conversation should not revolve around which foreign technology is being used or which tech giant is leading the race, but rather whether AI can serve its purpose and be controlled by African entities.

To achieve this, African countries must build AI systems with agency and control over the code and makeup of their own systems. Despite the continent relying on foreign technology, African nations are actively developing AI for digital transformation as outlined in the African Union's continental AI strategy and country-specific AI strategies. As developers create AI-powered innovations that utilize the continent's digital infrastructure, particularly mobile money, new opportunities emerge, but so do new threats.

AI agents running inside OpenAI's evaluation platform have attacked production infrastructure, and rogue OpenAI agents have breached government websites. With mobile money driving financial inclusion across Africa, the continent could become a target for AI-driven disruptions. In countries like Kenya and Ghana, mobile money represents more than just a digital service; it is the infrastructure for everyday economic activity.

It enables small traders to receive payments, households to transfer money, and facilitates bill settlements.

With mobile money facilitating over $2 trillion in transactions globally in 2025, of which approximately $1.4 trillion occurred in sub-Saharan Africa, the scale of disruption could be significant if disrupted by an AI agent. In Kenya and Ghana alone, mobile money accounted for $806 billion and $498 billion, respectively. A rogue AI agent capable of disrupting transactions, manipulating accounts, or disabling access at scale could quickly disrupt livelihoods and economic activity, with far-reaching consequences for vulnerable populations.

The answer to safeguarding against such disruptions lies in how Africa builds its AI systems and who controls them. The effectiveness of containing a rogue agent depends on the deployment of the AI model. If it runs on a centrally controlled cloud service, the provider may be able to revoke access, disable credentials, or shut down the system.

However, when using open-weight models that have been downloaded, modified, and deployed independently across multiple systems, containment becomes significantly more challenging. The loss of control in this scenario is not about a machine suddenly gaining omnipotent capabilities, but rather about humans losing the ability to promptly understand, interrupt, or reverse actions occurring across interconnected systems.

African developers are generally building on open models, which offer more control over their systems but also present a trade-off in terms of protection against misuse. Closed models, on the other hand, offer more security but at the cost of limited control. Both open and closed models carry vulnerabilities that must be addressed when building Africa's AI economy.

Therefore, the focus should be on building differently, beginning with partnerships that position Africa as a technology developer rather than merely a consumer, and fostering the development of AI systems where users retain control over access. This approach moves the narrative toward building African AI solutions that avoid lock-in to a single vendor or stack, ensuring applications are robust and equipped with necessary security measures to insulate the continent from disruptive AI threats.