Something has gone wrong in the way most institutions are thinking about AI and education, and it is going wrong quietly, at scale, before most policymakers have noticed.

The problem is not that schools and universities are using AI. AI in education is producing genuinely valuable outcomes: personalized learning pathways, early identification of students at risk of disengagement, administrative automation that returns hours to teachers, accessibility tools that reach learners who conventional formats were never designed to serve.

A genuine national AI education strategy requires confronting three challenges most strategy documents avoid: the vendor lock-in trap, the open-washing problem, and the question of what AI should actually develop in students.

The Procurement Trap That Nobody Planned For

The AI tools embedded in schools and universities across the world were not acquired through strategic national planning. They were acquired through the path of least resistance: a compelling demonstration, a free trial, a district-level pilot that became a system-wide dependency. The procurement trap works in four stages that are now well documented.

A vendor offers an AI platform at a subsidized or free entry price. The platform embeds itself into institutional workflow, student data flows into it, teachers build practice around it, assessments calibrate to it. The vendor then raises prices, changes data policies, modifies algorithms, or is acquired. The institution cannot exit because it would require rebuilding its entire educational technology stack.

The legislative response is accelerating. California’s Assembly Bill 1159, one of 134 active AI-in-education bills being tracked across 31 US states, specifically targets this gap by moving to prohibit ed-tech operators from monetizing student data.

The contrast between South Korea and New York City is instructive. South Korea committed nearly $740 million between 2024 and 2026 to train teachers in advanced AI tools, but when the National Assembly debated AI-powered digital textbooks, lawmakers voted to make adoption a school-level choice rather than a national mandate, understanding that tools deployed without governance are liabilities, not assets. New York City made the opposite bet. Parents packed a Panel for Educational Policy meeting in spring 2026 demanding the DOE pause all AI deployments while the city finalized its governance framework.

The procurement trap is not the result of bad intentions. It is the result of individually rational decisions made without adequate national frameworks that protect long-term interests.

What "Vendor Lock-In" Actually Means in Education

In enterprise contexts, vendor lock-in is primarily a financial and operational risk. In education, it carries three additional dimensions that make it qualitatively more serious.

Pedagogical lock-in. When AI systems determine how students receive feedback, how they are assessed, how learning gaps are identified, and how curricula are sequenced, the pedagogy is no longer defined by educators. It is defined by the model, optimized for outcomes the vendor could measure and monetize.

Epistemological lock-in. When a generation learns primarily through AI intermediaries built on particular models of knowledge and correctness, those models become an invisible layer of assumptions students lack the framework to question. UNESCO warns directly: when educational technologies are developed primarily for financial gain, education becomes a product and students become consumers, and deepening the AI divide.

Sovereignty lock-in. For countries in the Global South building national education systems on top of foreign AI platforms, the dependency is not just commercial. It is a form of technological colonialism: the foundations defining what counts as educated, what counts as correct, and what counts as capable are controlled by external interests. India’s response to this risk is architecturally significant, rather than building on foreign platforms, India constructed the National Digital Education Architecture (NDEAR), a federated, interoperable digital architecture built on open standards and open APIs, deliberately designed so that no single vendor can capture the stack.

The Open-Washing Problem

The response many institutions reach for when confronting vendor lock-in is “open-source AI.” If the model is open-source, the reasoning goes, there is no vendor dependency, no data extraction, no proprietary control. This reasoning is correct in principle and inadequate in practice, and the inadequacy has a name: open-washing.

Open-washing is the practice of marketing proprietary or semi-proprietary AI systems as open while maintaining tight control over their functioning and the critical elements of their technical ecosystem. Despite names that imply openness, many systems keep their essential internal mechanisms under tight commercial or governmental control.

Genuine openness in educational AI requires three things that open-washing does not provide. Open training data — educational AI models must be built on datasets that are auditable and contestable, free from the commercial interests that shape what counts as correct or worth knowing. Open governance — not a GitHub repository, but a decision-making architecture with genuine power for educators, students, parents, and communities. Open educational resources — the active participation of all educational actors in AI governance, treating education as a public good.

Estonia embodies this logic. The AI Leap initiative builds on X-Road, a public interoperability layer that has been open-standard and state-owned since 2001, equipping 20,000 high school students and 3,000 teachers with AI literacy grounded in civic purpose.

The Student Data Privacy Crisis

AI education platforms now collect unprecedented student data: keystroke patterns, eye movements, facial expressions, voice tone, learning speed, and emotional state indicators. This data describes not just what students know but how they think, how they struggle, and what environments they thrive in. It is among the most sensitive data in existence. And in most countries, it is largely unprotected.

Under the enforcement timelines of the EU AI Act, high-risk AI systems deployed in educational settings — those used for admissions, institutional access, student monitoring, or automated learning outcome assessments — must comply with mandatory transparency, human oversight, and conformity assessment requirements. The Act explicitly prohibits AI systems designed to infer the emotions of natural persons, addressing one of the most invasive categories of edtech surveillance.

Finland has built an answer that does not wait for regional enforcement. According to directives from the Ministry of Education and Culture, local education providers must conduct mandatory Data Protection Impact Assessments (DPIAs) prior to deploying any AI tools. Idaho, meanwhile, has enacted SB 1227, establishing a statewide K-12 framework banning AI from replacing human teachers entirely — a blunter instrument than Finland’s nuanced governance, but reflecting the same underlying recognition: without explicit legal constraints, commercial AI interests will govern by default.

What AI Should Actually Develop in Students

Here is the question most national AI education strategies fail to ask clearly: what, precisely, should students be learning about and through AI?

The dominant answer in most strategies is AI skills: prompt engineering, algorithmic thinking, data literacy, technical fluency. These are real and valuable. But they are incomplete, because they position AI as a tool to be operated rather than a system to be governed. The next generation will not primarily be AI engineers. They will be people who live and work in environments saturated with AI-assisted decisions — about credit, employment, healthcare, criminal justice, and education itself. What they need is not primarily the ability to use AI tools. It is the ability to evaluate, contest, and participate in governing them.

Four capabilities go well beyond tool proficiency:

What a Sovereign National AI Education Strategy Looks Like: Seven Countries

Different countries are arriving at different answers to the same question. The range of approaches is more instructive than any single model.

From Consumers to Architects

There is a generation of students entering education right now who will be the first to spend their entire academic lives in AI-saturated learning environments. What they learn about AI, and what they learn to do with it, to it, and deliberately without it, will determine whether they emerge as architects of AI-mediated society or subjects of it.

The countries generating the most instructive evidence did not treat AI education as a simple procurement choice. They treated it as a fundamental question about what kind of citizens their education systems are designed to produce. When institutions are left to define their own governance, university by university, the result is a dangerously fragmented landscape. When the state defines minimum governance standards — as Finland, Estonia, and India have done — institutions operate within a protective boundary that safeguards students while leaving room for educational innovation.

The UNICEF Digital Education Strategy, supporting initial rollout phases across 18 countries through Helsinki’s Global Learning Innovation Hub, demonstrates that this state-level governance approach can be shared across borders. Countries developing their initial institutional capacity can partner with those further along — a model of educational sovereignty that multiplies rather than divides.

The question for every institution, ministry, and policymaker making decisions about AI in education today is not “which AI tools should we adopt?” It is: whose interests does this adoption serve, and are they genuinely the same as our students’ interests?

The answer to that question, honestly given and structurally enforced, is the entire national AI education strategy.

Treating AI as a digital public good is not a utopian aspiration. It is the defining principle that distinguishes a sovereign national AI education strategy from a passive national vendor adoption plan.

Frequently Asked Questions

Common questions about national AI education strategies, vendor lock-in, and what institutions can do.