The promise of open-source artificial intelligence, heralded by companies like Stability AI, often arrives wrapped in the language of democratisation and accessibility. It suggests a future where innovation is shared, not hoarded, where the Global South can build its own digital destiny. Yet, here in Lesotho, a different, more opaque reality is taking shape. My investigation into Stability AI's quiet expansion across the continent, particularly its activities within our mountain kingdom, reveals a narrative far removed from the utopian visions of its public relations. What they're not telling you is that behind the veneer of collaborative development lies a strategic manoeuvre to secure invaluable data and influence, often at the expense of local control and transparency.
For months, whispers circulated through Maseru's tech hubs and government corridors about a foreign entity collecting vast datasets, ostensibly for AI training. Initially, these were dismissed as rumour, the usual anxieties surrounding foreign investment. However, my inquiries, prompted by a series of unusual data transfer requests observed by local network providers, began to paint a clearer picture. Sources close to the matter confirm that Stability AI, known for its Stable Diffusion image generation models, has been leveraging a network of ostensibly independent local contractors and NGOs to gather specific types of data. This data, ranging from anonymised demographic information to linguistic patterns in Sesotho, is then funnelled back to their global operations.
The official narrative, as presented by Stability AI in its limited public statements regarding African engagement, focuses on fostering local talent and contributing to digital literacy. "We believe in empowering creators worldwide," Emad Mostaque, Stability AI's former CEO, once stated, a sentiment echoed by current leadership. "Our open models are a catalyst for innovation, particularly in underserved regions." While such rhetoric is commendable, the practical application on the ground suggests a more extractive agenda. Documents obtained through a confidential source, including internal project proposals and contractor agreements, outline specific data acquisition targets that go far beyond what would be necessary for general model improvement. These targets include high-resolution satellite imagery of specific agricultural regions, detailed records of mobile money transactions, and even voice samples from various Basotho dialects.
One particular agreement, signed between a little-known local tech incubator, 'Thaba-Bosiu Digital,' and a Seychelles-registered shell company with indirect ties to Stability AI, details a multi-year project to digitise historical archives and local folklore. On the surface, this appears to be a benevolent cultural preservation effort. However, a closer examination of the contractual clauses reveals that 'Thaba-Bosiu Digital' effectively cedes all intellectual property rights and future commercialisation potential of the digitised data to the Seychelles entity. The financial compensation for this extensive work, while significant by local standards, pales in comparison to the potential value of such culturally rich and linguistically diverse datasets for training advanced AI models.
"This is not just about data collection, it's about knowledge extraction," remarked Dr. Puleng Mohlomi, a prominent Basotho linguist and advocate for digital rights at the National University of Lesotho. "Our languages, our stories, our very identities are being fed into these black boxes, and we have no say in how they are used, or for whose benefit. It is a new form of colonialism, digital colonialism, cloaked in the language of progress." Dr. Mohlomi's concerns resonate deeply within a nation that has long grappled with external influences shaping its destiny. The idea that Basotho cultural heritage could be monetised by foreign entities without equitable benefit or control is a bitter pill to swallow.
My investigation also uncovered a series of workshops and training programs, ostensibly designed to upskill local youth in AI development. While these initiatives are indeed valuable, the curriculum, as reviewed from leaked internal documents, heavily emphasises the use of Stability AI's proprietary platforms and tools. This creates a dependency, ensuring that any local innovation remains tethered to their ecosystem. It is a classic 'build the road, then charge the toll' strategy, expertly disguised as capacity building. The financial flows are equally telling. While Stability AI's public valuation has fluctuated significantly, reportedly reaching over $1 billion at one point, the funds allocated to these African initiatives are meticulously structured to avoid direct financial transparency or significant local equity participation. Follow the money, and you will find it flowing outwards, not circulating within.
This pattern is not unique to Lesotho. Similar arrangements have been observed in other African nations, where the allure of advanced technology and foreign investment often blinds local stakeholders to the long-term implications of data sovereignty. "The global race for AI dominance is fundamentally a race for data," stated Dr. Timnit Gebru, a leading voice in ethical AI research and founder of the Distributed AI Research Institute (dair). "Companies are desperate for diverse, high-quality datasets, and they will go to great lengths to acquire them, often exploiting existing power imbalances." Her insights provide crucial context to the operations I have uncovered here.
The implications for Lesotho are profound. As AI becomes increasingly integrated into daily life, from agricultural planning to public service delivery, control over the underlying data and models becomes paramount. If our foundational datasets are owned and controlled by foreign entities, our ability to shape our own digital future is severely compromised. It raises questions about national security, economic independence, and the very fabric of our cultural identity. The promise of open-source AI, in this context, risks becoming a Trojan horse, delivering external control disguised as liberation.
While Stability AI has faced its share of internal turmoil, including leadership changes and financial pressures, its strategic pursuit of global data assets remains unwavering. The company’s trajectory, from a darling of the open-source movement to a commercial entity grappling with profitability, highlights the inherent tension between idealism and startup reality. The pressure to generate revenue and justify valuations often overrides initial ethical considerations, particularly when operating in regions with less stringent data governance frameworks.
This is not to say that all foreign investment in AI is inherently exploitative. Indeed, genuine partnerships can bring immense benefits. However, the case of Stability AI in Lesotho serves as a stark reminder that vigilance is crucial. Governments, civil society organisations, and individual citizens must demand greater transparency, robust data protection laws, and equitable benefit-sharing mechanisms. As the digital landscape continues to evolve, the Basotho people, like many across Africa, must ensure they are not merely suppliers of raw data for the global AI machine but active, empowered participants in shaping its future. The future of our digital sovereignty depends on it.
For more on the broader implications of AI's expansion in Africa, consider reading NVIDIA's Golden Handcuffs: Is Africa Building on Sand, or a Solid Foundation? [blocked]. The challenges are interconnected, and the stakes are high. The conversation around AI governance must move beyond Silicon Valley boardrooms and into every corner of the globe, including our own mountain kingdom. The time for passive observation is over; the time for critical engagement is now. The future of AI, and indeed our future, depends on our collective ability to scrutinise these powerful forces. For further reading on the ethical considerations of AI, particularly in developing nations, one might consult resources from MIT Technology Review. The debate around data ownership and digital rights is only intensifying, as highlighted by various reports on Reuters Technology. We must be prepared to ask the difficult questions and demand accountability from those who seek to shape our digital world.
