The announcement seems innocuous: OpenAI Academy has started a collaboration with the Walton Family Foundation to organize “AI Skills Jams,” hands-on workshops for K–12 teachers. The stated goal is to provide practical AI skills to bring into the classroom. A piece of educational philanthropy, one might say. But reading between the lines, a design emerges that touches exposed nerves in the sector: whoever controls the platforms on which educators are trained controls how AI will be taught, used, and normalized in public schools.

The initiative fits a broader trend: large AI providers – OpenAI, Google, Microsoft – are investing heavily in digital literacy programs, often in partnership with foundations and local agencies. The immediate advantage is the penetration of proprietary tools into educational processes, creating a lock-in effect that starts right with teacher training. It’s not just about teaching how to use ChatGPT: it normalizes the idea that classroom AI must necessarily pass through closed, cloud-based interfaces governed by a single commercial entity.

For those closely following the debate on LLM adoption in education, this move raises second-order questions. The first concerns data sovereignty: school systems, especially in Europe under GDPR, should carefully assess whether teacher and student interactions with external tools guarantee full data control and local residency. The “practical” training proposed by OpenAI, if based on its commercial platform, risks creating dependencies without offering on-premise alternatives that public institutions could self-manage.

The second implication is less visible but equally structural: the very definition of AI competence. Promoting an exclusively applied use – prompt engineering, integration into teaching workflows – without delving into how models work, what their training entails, and what their intrinsic limits are, reduces teachers’ critical capacity. Instead of raising a generation of students able to evaluate and potentially challenge LLM outputs, we risk educating them toward passive acceptance.

OpenAI’s move can also be read as an indirect response to European projects that are experimenting with sovereign cloud infrastructures and open-source stacks for education. While on one hand the supply of locally executable models grows – thanks to advances in quantization and on-device inference frameworks – on the other, Big Tech accelerates penetration into the places where tomorrow’s decision-makers are trained. By doing so, they raise the political cost of any future switch to self-hosted alternatives, because the entire body of teaching materials and teacher skills will be intertwined with specific tools.

This is not about demonizing OpenAI, nor about rejecting AI in schools. But the initiative, however generous its intentions, opens a front that must be carefully guarded: the training of trainers is the most sensitive point in the chain. Whoever influences it today determines tomorrow’s architecture.