Last year, Sam Altman invited Dave Eggers to speak to roughly 200 OpenAI staff members. It’s the kind of invitation that usually comes with a certain expectation: a celebrated writer and public figure known for building institutions around storytelling and education would arrive with a message about creativity, craft, or the future of writing—something inspirational, maybe even practical.
Instead, Eggers reportedly used the talk to deliver a warning that landed like a rebuke. According to the Financial Times, he told OpenAI employees that ChatGPT’s effect on educators’ lives is “catastrophic,” arguing that even if the technology wasn’t designed to do harm, it has nonetheless changed the conditions under which teaching happens. In his framing, the result isn’t merely inconvenience or disruption; it’s a deeper cultural shift—one that could silence an entire generation, not by censoring anyone directly, but by altering incentives, workflows, and expectations until educators and students stop doing the kinds of work that build literacy, voice, and confidence.
The Verge later summarized the episode, emphasizing Eggers’ bluntness and the unusual angle of the message: rather than focusing on what AI can do, he focused on what it does to people who teach and to the systems that shape learning. The story matters not only because Eggers is a prominent author, but because his critique points to a question that has been hovering around generative AI since its mainstream arrival: what happens when the tools become so capable—and so easy—that they quietly rewrite the definition of “effort,” “authorship,” and “learning” itself?
To understand why Eggers’ comments resonated inside a company built around language models, it helps to look at the specific audience. OpenAI staff are not just consumers of AI; they are the people designing the interfaces, shaping the safety policies, choosing what the model optimizes for, and deciding how the product behaves in real-world settings. A writer’s critique aimed at educators is therefore not a distant moral argument. It’s a feedback loop: it suggests that the product’s impact is not limited to novelty or productivity gains, but extends into the daily labor of teaching—where small changes in practice can compound into large changes in outcomes.
Eggers’ reported emphasis on educators is also significant because teachers are often treated as an afterthought in AI discussions. Many early conversations centered on students, on academic integrity, or on the fear that AI would replace student writing. But educators are the ones who translate policy into practice. They decide what counts as evidence of learning, how assignments are structured, how feedback is delivered, and what kinds of mistakes are allowed to happen. If AI changes the environment in which those decisions are made, then the effects are likely to be systemic rather than isolated.
In the classroom, the shift can be subtle at first. A student submits an essay that reads smoothly. The teacher can’t easily tell whether the student wrote it, revised it, or merely prompted it. Even when the teacher suspects AI involvement, the question becomes: what now? Do you redesign assignments? Do you require drafts? Do you move toward oral defenses? Do you grade process instead of product? Each option demands time, training, and institutional support—resources that many schools already struggle to provide.
And that’s where Eggers’ “catastrophic” framing becomes more understandable. The harm he describes isn’t necessarily that every student uses AI to cheat. It’s that the presence of AI changes the baseline. It forces educators to spend more time policing uncertainty, more time rethinking assessment, and more time dealing with the emotional and ethical fallout of a tool that can produce convincing text on demand. When teachers are forced into constant triage—trying to determine what is authentic, what is learned, and what is generated—the job becomes harder in ways that don’t always show up in headlines.
There’s another layer, too: the psychological effect of AI on both teachers and students. For students, writing is often a fragile process. It involves risk—risk of being wrong, risk of sounding awkward, risk of not knowing how to start. If AI can generate a polished draft instantly, the risk shifts. Students may feel less permission to struggle, less motivation to develop their own voice, and less comfort with the messy middle of learning. For teachers, the risk shifts as well. They may feel that their expertise is being undermined by a tool that can mimic competence. Even when teachers are skilled at reading for nuance, AI can blur the signals they rely on.
Eggers’ critique, as reported, suggests that these shifts aren’t temporary. They become normalized. Over time, the classroom adapts to the tool rather than the tool adapting to the classroom. That’s a crucial distinction. If AI is treated as a shortcut that students will use regardless of instruction, then educators are pushed toward strategies that assume AI is present. Those strategies can be effective, but they also change what education looks like. They can narrow the range of tasks that are feasible to assess, and they can reduce the space for open-ended writing—precisely the kind of writing that helps students learn how to think.
This is where the phrase “silencing an entire generation” becomes more than rhetoric. Silence doesn’t have to mean literal suppression. It can mean that fewer people speak in the ways they otherwise would have. If students stop writing because they believe their writing won’t matter—or because they believe AI can do it better—then the long-term effect is a reduction in practice. And practice is the engine of skill development. Writing, especially, is not just a performance; it’s a cognitive process. When the process is outsourced, the learning that depends on it can weaken.
But there’s a counterargument that often appears in these debates: AI can also help students who struggle. It can provide scaffolding, translation, brainstorming, and feedback. It can lower barriers for learners with disabilities or language differences. It can help teachers manage workload by drafting lesson materials or generating examples. In other words, the same tool that can enable shortcuts can also enable support.
Eggers’ reported position doesn’t necessarily deny those benefits. His claim, as described, is about outcomes and environment. Even if AI offers assistance, it can still create conditions where the assistance becomes the default. When a tool is available everywhere, at low cost, and produces fluent text quickly, it becomes tempting to treat it as the solution rather than the supplement. The question becomes: who controls the use? Who sets the norms? Who ensures that the educational value remains intact?
That’s why the conversation inside OpenAI—if Eggers’ message was taken seriously—would likely have touched on product design and policy. Generative AI systems don’t just output text; they shape behavior through their affordances. If the product makes it easy to generate essays, summarize readings, and rewrite drafts, then it will naturally be used for those purposes. If the product is integrated into school workflows, it will become part of the routine. And once it becomes routine, it becomes hard to separate “help” from “replacement.”
One unique aspect of this story is that it comes from a writer who has spent decades building platforms for writers and arts communities. Eggers is not a technologist arguing from abstraction. He’s someone who has watched how publishing, journalism, and education evolve under economic pressure and technological change. His concern, as reported, is that AI is not simply adding a new instrument to the creative toolkit—it’s changing the incentives and expectations around authorship itself.
Authorship is a loaded word in the age of AI. When text can be generated instantly, the meaning of “writing” shifts. Is writing the act of producing words, or is it the act of thinking, organizing, revising, and taking responsibility for claims? In education, those distinctions matter. Teachers want to know what students can do, not just what they can submit. But if the submission becomes decoupled from the process, assessment becomes less reliable. And when assessment becomes less reliable, institutions either tighten enforcement—which can be punitive and time-consuming—or they lower standards—which can be demoralizing and harmful.
Eggers’ warning, then, can be read as a plea for the people building these systems to consider downstream effects earlier and more seriously. Not just whether the model can produce text, but whether the product encourages behaviors that undermine learning. Not just whether the system is safe in a technical sense, but whether it is safe in a social sense—safe for the institutions that shape young people’s development.
There’s also a broader cultural issue at play: the speed of adoption. Generative AI arrived with a level of capability that outpaced the ability of schools, universities, and policymakers to respond. That mismatch creates a vacuum. Into that vacuum rushes whatever is easiest. If AI makes it easier to produce text than to learn how to produce it, then the vacuum fills with shortcuts. Even well-intentioned educators can end up fighting a losing battle if they’re asked to enforce norms that the technology makes obsolete.
This is why the story is not only about OpenAI or about one author’s opinion. It’s about governance and responsibility across the ecosystem. Schools need guidance on acceptable use. Universities need assessment models that measure learning rather than output. Policymakers need frameworks that balance innovation with protection of educational integrity. And companies need to consider how their products will be used, not just how they are marketed.
A unique take on Eggers’ reported comments is to treat them as a signal about the “invisible curriculum” of AI. The visible curriculum is what schools teach: writing skills, critical thinking, research methods, argumentation. The invisible curriculum is what students learn from the environment: what gets rewarded, what gets punished, what is expected, what is considered legitimate. If AI changes what is rewarded—polished output over rough drafts, speed over iteration, fluency over originality—then the invisible curriculum shifts. Students absorb those lessons even if teachers never explicitly teach them.
In that sense, “silencing” could refer to a narrowing of the kinds of learning that are valued. When AI makes