The tech industry’s relationship with the creative and knowledge sectors has long been described in terms of disruption, transformation, and, more recently, extraction. But a newly unredacted legal filing has stripped away the euphemisms, revealing that leaders at Microsoft and OpenAI themselves have privately characterized their generative artificial intelligence models with a word far more damning: theft. The documents, unsealed in a New York federal court, contain internal communications where executives describe the technology as, in one stark phrasing, the “largest theft of labor in human history.”
Internal Language Exposes a Raw Assessment of GenAI
The filings, part of an ongoing lawsuit, offer an uncommonly candid window into how the architects of the most powerful large language models (LLMs) view their own work. OpenAI’s policy director, Jack Clark, wrote that the company was “creating systems that substitute for the labor of the people that define the ‘culture’ of society.” This is not a statement from a detached critic, but from an insider describing the product’s core economic logic.
Microsoft’s internal policy documents were even more blunt. They asserted that generative AI could “significantly disrupt the employment of the very people who generated the data on which the foundation model was trained.” The document continues with a chillingly precise description of the business model: “LLMs are a product that destroys its supply chain.” In other words, the raw material of these models—the articles, books, code, and art created by human beings—is not simply used; it is consumed in a process that, at scale, eliminates the need for the creators themselves.
This admission from Microsoft and OpenAI confirms what many journalists, authors, and artists have long suspected: that the economic promise of generative AI is built directly on the devaluation of human-generated content. The internal language reframes the public debate. It is no longer a question of hypothetical disruption. The companies themselves have acknowledged that their products, by design, render their own source of training data economically unviable.
The Destroyed Supply Chain: What the Filing Reveals About Journalism
The legal documents provide specific evidence regarding the threat posed to news publishing. OpenAI’s own internal assessment labeled the company an “existential threat” to news organizations. This acknowledgment is particularly striking given the public, multi-million dollar partnerships OpenAI has pursued with major publishers. The unredacted filings suggest these deals may be less about building a sustainable ecosystem for journalism and more about securing access to high-quality training data before the supply chain collapses entirely.
A deposition from an OpenAI software engineer further clarified the dynamics of web traffic, the lifeblood of digital publishing. The engineer testified that “no matter how prominently we show the links, users won’t click.” This single statement dismantles the industry’s prevailing narrative that AI-generated search results and summaries will drive discovery and traffic to original sources. The engineer’s testimony indicates that a generation of users, when given a complete answer generated by an LLM, has no incentive to visit the original article. The link, even when prominently displayed, becomes a vestigial organ.
This has immediate, concrete implications for anyone building a business on writing. The economic model of advertising-supported journalism relies on page views. If generative AI provides the answer without the click, the advertising revenue disappears, and the incentive to produce the original reporting vanishes.
How Does the “Theft” Actually Work at Scale?
To understand the scale of the problem, one must understand the mechanics of training a large language model. These models do not “read” articles the way a human does. They process vast quantities of text—billions of words scraped from the public internet, including copyrighted news articles, books, and academic papers. During training, the model identifies statistical patterns in how words are used together. It does not store a copy of an article, but it learns the patterns that article contains.
The critical issue is that this process occurs without the permission of the copyright holders, and until recently, without compensation. When the model is later asked a question, it uses these learned patterns to generate a statistically plausible response. For factual questions, this response often closely resembles the phrasing and information contained in the specific articles it was trained on. The filing reveals that Microsoft and OpenAI recognized this not as transformative fair use, but as the substitution of human labor—a direct replacement of the journalist by the machine.
The phrase “destroys its supply chain” perfectly captures the vicious cycle. A human writer produces a detailed article. A company like OpenAI trains a model on that article and thousands like it. The model then generates answers that satisfy a user’s query, eliminating the need for the user to visit the original article. Without traffic, the publication loses revenue. Without revenue, the publication cannot afford to employ the writer. The writer stops producing the article. The supply of high-quality, original data dries up. The model, unable to learn from new human work, becomes stagnant and increasingly prone to error as it is trained on its own previous output—a process known as model collapse.
A Growing Ecosystem of Stochastic Gruel
The problems identified in the Microsoft and OpenAI filings are not limited to the actions of these two companies. A parallel ecosystem has emerged in which the technology is used to create vast quantities of low-quality content specifically designed to capture advertising revenue. A recent investigation detailed the operations of a media firm that buys established journalism outlets and systematically converts them into what one critic described as “lakes of stochastic gruel.”
These operations follow a playbook that exploits the very structure of the internet. They publish enormous volumes of AI-generated articles that are grammatically sound but factually empty, designed to rank in search engines. The investigation noted that this strategy enjoys “enormous success at capturing reader eyes and attention through platforms that actual journalism relies on—while publishing oceans of trash that would get a real journalist fired immediately.” The firm’s responses to questions about its practices were described as “empty and contradictory platitudes about transparency, accountability, and responsible AI use.”
This is the environment that the unredacted filings illuminate. It is not a distant hypothetical. It is the current reality of the information economy. A major technology company has admitted internally that its flagship product is designed to destroy the industry that provides its foundational material, while a network of smaller firms uses the same technology to flood the digital commons with noise.
The Broader Cultural and Economic Implications
The language used by Jack Clark at OpenAI is particularly revealing. He described the systems as substituting for the labor of the people who “define the culture of society.” This moves the discussion beyond journalism and into the realm of cultural production as a whole. If the labor of writers, artists, musicians, and filmmakers is the raw material for these systems, and if those systems subsequently replace the creators, the long-term cultural consequence is a flattening of creative output.
Humans write, paint, and compose from a combination of lived experience, emotion, and a unique historical perspective. A language model has no experience, no emotion, and no perspective. It has only a statistical map of what has already been said. If the economic incentive to produce new, original cultural work is removed because that work is immediately absorbed and devalued by an AI, the cultural conversation becomes a closed loop of recombined past data. The output becomes increasingly derivative, a mirror of a mirror with no new light being shined on it.
The internal documents suggest that the executives building these systems are acutely aware of this dynamic. They have not stumbled upon it. They have identified it as the core feature of their product.
What This Means for the Individual Creator
For a freelance writer, a game journalist, or a novelist, the implications of these filings are deeply personal. The promises of the technology’s advocates—increased productivity, democratized access to information, a new era of creativity—ring hollow when set against the cold calculus of the destroyed supply chain. The individual creator is no longer competing against another creator with a better idea. They are competing against a system that has ingested their entire corpus of work and can now reproduce its patterns for free.
The path forward is uncertain. Legal battles over copyright and fair use will continue for years. New compensation models, such as collective licensing, are being proposed but face significant logistical hurdles. The most immediate and reliable strategy for an individual creator remains the cultivation of a direct relationship with an audience. Substacks, podcasts, Patreon-supported communities, and live events represent a return to a patronage model where the audience pays directly for the work, bypassing the advertising-driven platforms that are being hollowed out by AI-generated content.
The badger may have taken the cat’s dinner, but the cat—plump, glossy, disdainful of the front door—still has its own resources. So too must the writer. The data supply chain is under attack, but the human need for a narrative, for a perspective, for a story told from a specific point of view, has not changed.
A New Lens for Understanding the Technology
The unredacted court filings do not simply provide evidence for a legal argument. They provide a lexicon. The phrase “destroy its supply chain” should replace the vague, optimistic language of “disruption” in every business meeting, every boardroom, and every public relations statement. It is a more accurate descriptor of the mechanism at work.
When next a tech executive speaks of “democratizing creativity,” one can now look at the internal documents of the largest companies in the sector. What they found was not a beautiful vision of a post-scarcity creative utopia. What they found was a cold, clear-eyed admission that the business plan is to vacuum up the work of the present to build a product that eliminates the need for the future. It is the largest theft of labor in human history, and the perpetrators have, in a moment of legal necessity, signed their own confession.
