Language inclusion is not sovereignty: the uncomfortable question behind Google’s AI update

Adding Afrikaans, isiZulu, Sesotho and Setswana is progress, yet it also spotlights who controls the infrastructure, training data, and economics of African-language AI.

Article link: https://open.substack.com/pub/johanosteyn/p/language-inclusion-is-not-sovereignty?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Audio summary: https://youtu.be/e17HoYyAd2c

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

When Google adds four South African languages to its AI search experiences, it feels like a rare moment of genuine inclusion. Afrikaans, isiZulu, Sesotho and Setswana are now supported in Google’s AI Overviews and AI Mode, which means more people can search and learn in the language that feels most natural to them. That matters. But there is a second, less comfortable conversation we need to have: language inclusion is not the same as sovereignty. 

A platform can “support” our languages while the underlying ownership of data, models, infrastructure, and economics remains elsewhere. If South Africa wants AI that truly serves us, we need to think beyond access and ask who controls the foundations.

CONTEXT AND BACKGROUND

Google’s update is part of a broader rollout across Africa. In South Africa, the headline is the addition of four local languages, framed as a step towards ensuring people are not excluded from the “AI economy” because English is not their first language. Google also pointed to its WAXAL language work as part of the foundation for this expansion.

WAXAL is worth understanding because it points to a more hopeful model. It is presented as a large-scale African language speech corpus, built with partnerships involving African academic and community organisations, and released openly for research and development. Its accompanying research paper describes the dataset, its methodology, and its limitations and ethical considerations.

There is a wider continental backdrop, too. In January 2026, the Pan-African Parliament publicly called for African sovereignty over sensitive data and the AI systems built on it, warning about a new era of data colonialism if ownership and control are not addressed.

INSIGHT AND ANALYSIS

Here is the key tension. Language support improves access. Sovereignty is about control and benefit-sharing. The first can happen without the second. A multinational can localise interfaces and outputs while the value created by the system, and the rules governing it, remain external.

This matters because language is not just vocabulary. Language carries culture, context, and power. When AI systems summarise information, answer questions, and shape what people see, they are influencing knowledge and decision-making at scale. If the underlying data pipelines and model tuning do not properly reflect local realities, AI can confidently flatten nuance. Worse, when the incentives are commercial, the system will prioritise whatever is easiest to scale, not necessarily what is most culturally faithful or socially beneficial.

What I find encouraging is that WAXAL hints at a different bargain: one where African institutions have a meaningful role in collecting, stewarding, and shaping language resources. Nigerian reporting on WAXAL emphasised that African universities and community organisations led parts of the effort, supporting the idea that the continent not only supplies raw data, but can grow research capability too.

Yet even open datasets raise uncomfortable questions. “Open” is a gift to researchers and start-ups, but it can also become a one-way pipeline if global firms take the value and local communities see little return. This is why the conversation is shifting from “more data” to “better deals”: consent, governance, and fair value exchange.

We are seeing experiments in this direction. In November 2025, Axios reported on Mozilla’s attempt to establish a data-sharing collective designed to give communities more control over how their data is used for AI, and to explore fairer models of participation.

IMPLICATIONS

For South African policymakers, this is not a niche technology issue. It is an industrial strategy. If our languages become part of the AI layer of the internet, we should be building local capability in data stewardship, evaluation, and governance, not only celebrating feature rollouts. This includes clear rules for sensitive data, but also incentives for local research, local start-ups, and public-private partnerships that keep skills and value in the country.

For business leaders, language AI is a productivity opportunity, but also a risk surface. If employees start relying on AI summaries and AI search in local languages, organisations need guidance on verification, source quality, and appropriate use. Trust is built through process, not optimism.

For educators and parents, the promise is obvious: learners can ask questions in the language that feels most accessible. The responsibility is equally clear: children must learn that AI can be wrong, and that language fluency does not equal truth.

CLOSING TAKEAWAY

Google adding Afrikaans, isiZulu, Sesotho and Setswana is a welcome step towards inclusion. But South Africa should not confuse inclusion with sovereignty. The deeper question is who will own the language resources, the evaluation standards, the governance rules, and the long-term economic upside. If we get this right, local-language AI can expand access to knowledge and opportunity. If we get it wrong, we risk a new form of dependency, where our languages enrich global systems while our ability to shape them remains limited. The next phase must be about capability, stewardship, and fair value, not just features.

Author Bio: Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net