South Korea Bet $850 Million on AI Textbooks and Lost. Here's Why Every Nation Should Pay Attention.

The government had committed more than 1.2 trillion won ($850 million) to the project. South Korea's largest EdTech companies had restructured their entire product roadmaps around it. Teachers across the country had been trained, tablets had been shipped, and a new digital curriculum had been written from scratch. And then, four months after the first AI-powered textbooks hit classrooms, the National Assembly voted to strip them of their legal status as official teaching materials. The $850 million dream had become an $850 million cautionary tale , and nobody in Silicon Valley, Brussels, or Beijing seems to be taking notes.

What Actually Happened

In June 2023, then-President Yoon Suk Yeol's administration unveiled the AI Digital Textbook Promotion Plan with a bold vision: by 2028, every core subject in Korean elementary, middle, and high schools would be delivered through AI-powered tablets. The initiative was conceived as both an educational upgrade and a statement of national technological ambition. South Korea, already a global leader in broadband penetration and digital infrastructure, would be the first country in the world to replace traditional textbooks with AI systems at national scale.

The government moved with unusual speed. Over the following eighteen months, more than 1.2 trillion won ($850 million) was allocated , covering device procurement, content development, teacher training, and infrastructure upgrades. Dozens of domestic EdTech firms signed contracts expecting mandatory government-backed adoption to follow. When the first semester rollout began in early 2025, approximately 37% of eligible schools had adopted the system. That number collapsed to 19% by the second semester starting September 2025 , a drop that preceded the National Assembly's formal vote on August 4 to terminate the program entirely, stripping AI textbooks of their legal status as official teaching materials in South Korea.

Why This Matters More Than People Think

On the surface, this looks like a story about a bad technology rollout , rushed procurement, poor testing, a government that moved too fast. And in one sense, it is. The AI textbooks that landed in classrooms were riddled with factual errors. Students encountered wrong answers presented with algorithmic confidence. Teachers, rather than having their workloads reduced as promised, found themselves spending additional hours fact-checking AI-generated content before presenting it to classes. The system also created a data privacy burden: parents raised legitimate concerns after discovering that the platforms logged detailed behavioral and attention data on minors without sufficient disclosure.

But the deeper significance extends far beyond South Korea's classroom walls. This was the most ambitious state-backed AI deployment in education in human history , and it failed at the most fundamental level: the AI was not accurate enough to be trusted as a primary educational source. At a moment when governments worldwide are racing to embed AI into national curricula, South Korea's collapse is not an edge case. It is a preview. Every nation currently drafting AI-in-education frameworks should study this failure not as a distant cautionary tale but as the best available real-world data point on what happens when AI deployment timelines are driven by political ambition rather than pedagogical readiness.

The Competitive Landscape

South Korea's AI textbook collapse lands at an awkward moment for global EdTech. Hundreds of companies , from US giants like Google Classroom and Khan Academy to Chinese adaptive learning platforms like Squirrel AI , have been pitching AI-powered learning systems to governments and school districts on the premise that personalized AI tutoring can replace or supplement traditional instruction. The South Korean implosion creates a new sales objection that will be very difficult to answer: what happens when the AI is wrong, and a child learns it?

The collapse also reshuffles the competitive landscape among South Korean EdTech firms. Companies that had concentrated their R&D budgets and product roadmaps on AI textbook content now face existential uncertainty, with reports of several companies on the verge of collapse following the government's reversal. Meanwhile, the traditional hagwon (cram school) industry , a $20 billion sector in South Korea that many had believed the AI textbook program would disrupt , has emerged intact and arguably strengthened. The perceived failure of AI education may reinforce the argument that human tutors remain irreplaceable at this stage of the technology's development. Distressed Korean EdTech assets are likely to attract acquisition interest from international platforms over the next twelve months.

Hidden Insight: The Confidence Problem AI Has Never Had to Confront at Scale

There is a dimension to this story that virtually every tech industry analysis has chosen to ignore: the specific failure mode was not a lack of raw capability, but a lack of calibration. The AI textbooks were generating confident wrong answers. That is qualitatively different from a system that flags uncertainty or declines to answer. A student who encounters a wrong answer delivered with the authoritative tone of a printed textbook is not just uncorrected , they may be harder to correct later. Learning science research has long documented that confident misinformation is substantially more resistant to correction than acknowledged uncertainty. AI systems optimized for fluency and coherence are particularly prone to this failure in instructional contexts , and the educational consequences are more durable than in almost any other AI application domain.

This points to a second-order problem for AI deployment broadly: the mismatch between how AI systems communicate certainty and the actual reliability of their outputs. In consumer applications, a hallucinated restaurant recommendation is an inconvenience. In a classroom, it is a curriculum artifact that a child may carry into an exam, a career, or a lifetime belief. The South Korean program discovered, at scale and at enormous cost, that the calibration requirements for educational AI are substantially higher than for commercial AI , and that the industry was not close to meeting them in 2025.

There is also a political economy lesson that deserves more attention. The AI textbook program was championed by President Yoon Suk Yeol, who was subsequently impeached and removed from office in early 2025. The program's fate became entangled with the political collapse of its patron, accelerating its termination beyond what pure technical failure alone would have warranted. This pattern , AI initiatives that become associated with specific political actors and therefore share their political fortunes , is a deeply underappreciated risk factor for large public-sector AI deployments. Governments designing AI education programs should build institutional independence robust enough to survive leadership changes. Programs that cannot outlast their political champions will consistently fail before they can prove themselves, regardless of the underlying technology's merit.

What to Watch Next

The most important indicator is how other countries respond to South Korea's failure. As of May 2026, the United Kingdom, Singapore, Australia, and the UAE all have active AI-in-education initiatives at various stages of development. If their ministries begin quietly adding accuracy benchmarks, hallucination rate requirements, and third-party calibration audits to AI procurement specifications , criteria that did not exist in South Korea's program , that signals the industry has absorbed the lesson. If instead they press forward on unchanged timeline assumptions, expect analogous failures within 24 to 36 months. The window for the global EdTech industry to self-regulate before governments mandate standards is probably no longer than 18 months.

Watch also for the South Korean EdTech fallout and subsequent repositioning. Companies that survive this episode may pivot to B2B enterprise training markets, where adult learners tolerate error more readily and where AI tutoring can be framed as a supplement to human instruction rather than a replacement. Several Korean EdTech firms already have significant international reach through language learning platforms. The survivors who successfully reposition will offer a useful model for how AI education companies can navigate the gap between today's capability and the institutional bar that schools, governments, and parents actually require. Expect cross-border M&A activity in Korean EdTech within the next 12 months as international platforms look to acquire distressed assets and local expertise at depressed valuations.

South Korea spent $850 million discovering what the rest of the world has not yet admitted: an AI that is confidently wrong is more dangerous in a classroom than one that is honestly unsure.

---