A School Used AI to Pull 200 Books From Its Library. The Fallout Says a Lot About How We're Letting Algorithms Shape Education.
When a secondary school in Greater Manchester fed its library catalog through an AI system to flag "inappropriate" titles, the result was a 200-book removal list that included George Orwell's 1984 and Stephenie Meyer's Twilight. The irony of an algorithm banning the most famous novel about authoritarian thought control barely needs stating. But the incident — which LBC reported this week — raises a question that extends well beyond one school's shelves: what happens to students' ability to think critically when the institutions meant to develop that skill outsource judgment to machines?
This isn't just about book bans. It's about a growing pattern in which AI systems, shaped by their training data, their guardrails, and sometimes by deliberate government censorship — are quietly narrowing the information students encounter. And the consequences may be harder to reverse than a bad reading list.
The Greater Manchester Incident
The school, which LBC did not name, tasked senior staff with reviewing its library collection. Rather than relying on librarians or educators to assess each title, they used an AI tool to earmark roughly 200 books for removal, flagging them as inappropriate for students. The list reportedly included Orwell's 1984 and Twilight, among others.
The school's librarian was, in LBC's words, "gobsmacked." And the reaction is understandable. Librarians spend years developing expertise in age-appropriate collections, reading culture, and the pedagogical value of challenging texts. Replacing that judgment with an algorithmic sweep isn't efficiency — it's abdication.
What makes this case instructive is its ordinariness. This wasn't a politically motivated purge or a culture-war flashpoint. It was a school trying to save time. That mundanity is precisely what makes it worrying. If AI-driven content filtering becomes a default administrative shortcut, the cumulative effect on what students can access, read, and debate could be enormous — and largely invisible.
The AI tool's specific methodology hasn't been publicly detailed. But the outcome reveals a familiar problem: content moderation systems, whether designed for social media or school libraries, tend to be blunt instruments. They struggle with context. A novel that depicts violence to critique totalitarianism gets the same flag as one that glorifies it. A book containing sexual content aimed at helping teenagers understand relationships gets treated the same as explicit material with no educational value.
How Censorship Gets Baked Into AI
The Greater Manchester case involved an AI tool making local decisions about a library. But the deeper issue is structural. The large language models and AI systems increasingly used in education don't arrive as blank slates. They carry the biases, omissions, and censorship patterns embedded in their training data.
Research from UC San Diego demonstrated this concretely. Political science professor Margaret Roberts and PhD student Eddie Yang examined AI language algorithms trained on two different Chinese-language text sources: the Chinese-language Wikipedia, which is blocked inside China, and Baidu Baike, a similar encyclopedia operated by Baidu that is subject to government censorship.
The findings were striking. The type of language algorithm they studied learns by analyzing how words appear together across large bodies of text, representing different words as connected nodes in a kind of semantic space — the closer the nodes, the more similar the meaning. When trained on the censored Baidu Baike corpus, the algorithm learned to associate the word "democracy" with concepts closer to "chaos" or instability.
Trained on the uncensored Wikipedia data, "democracy" clustered near "stability" and positive governance concepts.
This matters because these word associations aren't abstract. They flow downstream into chatbots, translation tools, autocomplete features, and educational software. A student using an AI tutor built on subtly censored training data might receive explanations of political concepts that reflect the biases of a particular regime — without either the student or the teacher realizing it.
Chatbots as a New Censorship Frontier
The problem isn't hypothetical. A 2023 WIRED analysis by Freedom House researchers found that authoritarian governments have moved aggressively to control what AI chatbots can say. The piece noted a telling contrast: ask ChatGPT what happened in China in 1989, and it describes the Tiananmen Square massacre. Ask the same question to Ernie, Baidu's AI chatbot, and it responds that it does not have "relevant information."
Freedom House researchers Allie Funk, Adrian Shahbaz, and Kian Vesteinsson — coauthors of the organization's Freedom on the Net 2023 report — described this as an early warning about a new frontier of online censorship. When OpenAI, Meta, Google, and Anthropic initially made their chatbots widely available, millions of users in countries with restricted internet access used them to access unfiltered information. For the roughly 70 percent of the world's internet users who live in places where governments block social media platforms, independent news, or content about human rights and LGBTQ issues, these tools briefly offered a workaround
Authoritarian regimes noticed. China and Russia, among others, have moved to develop domestic AI chatbots with built-in content restrictions, or to regulate how foreign chatbots operate within their borders. The result is a splintering of the AI information landscape along political lines — a dynamic that has direct implications for education.
The Classroom Connection
When students in different countries ask the same AI assistant the same historical question and get fundamentally different answers — one factual, one evasive — the technology isn't just reflecting censorship. It's amplifying and automating it at scale, with a veneer of objectivity that makes it harder to detect.
This is qualitatively different from a biased textbook. A student can compare textbooks, and a teacher can contextualize a flawed one. AI tools present themselves as neutral information sources. They don't come with author names, publisher reputations, or editorial stances that a critical reader can evaluate. They just answer.
What This Means for Critical Thinking
The core function of education — especially in the humanities and social sciences — is to expose students to complexity, disagreement, and discomfort. Critical thinking develops when students encounter ideas that challenge their assumptions and learn to evaluate evidence, weigh competing arguments, and form their own conclusions.
AI-driven content filtering threatens this process in two ways.
First, it narrows the input. When an algorithm removes 1984 from a library shelf or a chatbot declines to discuss a historical event, the student never encounters the material in the first place. You can't think critically about information you've never seen. The Greater Manchester case resulted in the removal of titles that are widely considered essential to developing literary and political awareness. An AI system that flags Orwell as inappropriate is one that fundamentally misunderstands the purpose of education.
Second, it flattens the process. When students interact with AI tools that present sanitized, consensus-driven responses, they lose practice with the messy work of intellectual inquiry. If every answer is smooth and confident, students don't learn to question sources, identify gaps, or recognize when information has been shaped by external pressures.
The UC San Diego research shows that these biases can be subtle — embedded in word associations rather than explicit omissions. A student using an AI writing assistant might never notice that the tool consistently frames certain political concepts in negative terms. The censorship is invisible precisely because it operates at the level of language patterns rather than obvious redactions.
What Should Happen Next
The solution isn't to ban AI from educational settings. These tools offer genuine benefits for personalized learning, administrative efficiency, and accessibility. The solution is to insist that AI remains a tool, not a decision-maker, when it comes to intellectual content.
Several principles should guide how schools, districts, and education ministries approach this:
Human oversight must be non-negotiable. AI can help organize, recommend, and flag — but the final decision about what students read and learn should rest with trained educators. The Greater Manchester librarian's expertise was the check that caught the algorithm's errors. Not every school will be lucky enough to have that check in place.
Transparency about training data matters. Schools adopting AI tools should demand clear documentation of what data the systems were trained on and what content policies govern their outputs. The Freedom House research covered by WIRED makes clear that different AI systems can produce radically different responses to the same questions depending on their origins and restrictions. Educators need to understand those differences.
Students should learn how AI filtering works. Rather than shielding students from the reality of algorithmic bias, schools should teach it. Understanding how AI systems can embed and propagate censorship is itself a critical thinking skill — arguably one of the most important for the coming decades.
Diverse sources must be preserved. Libraries, both physical and digital, exist to provide access to a range of perspectives. AI tools that collapse that range into a single "appropriate" output undermine the fundamental purpose of educational collections.
The Bigger Picture
The Greater Manchester incident is small in scale — one school, 200 books, a decision that may well be reversed after public attention. But it sits at the intersection of several larger trends: the rapid adoption of AI in education, the global fragmentation of AI systems along political and cultural lines, and the growing tension between efficiency and intellectual freedom.
Orwell wrote 1984 as a warning about the machinery of thought control. The machinery he imagined was crude — telescreens, thought police, the physical rewriting of history. The version emerging now is subtler. It's an algorithm that quietly decides what's appropriate, a chatbot that declines to discuss inconvenient history, a language model that learned from censored text and carries that censorship forward into every interaction.
The students who never read 1984 because an AI flagged it won't know what they missed. That's the point. And that's the problem.
