.png)
Authors Push Back on AI Fair Use: The OpenAI Copyright Fight Reaches a Defining Moment
- Rob Frankson
- 24 hours ago
- 4 min read
AI & The Author — 7 September 2026
Artificial intelligence, writing, publishing and what the latest developments mean for working authors.
Quick Shortcuts
The copyright argument has reached a sharper moment
The argument over whether artificial-intelligence companies can train their systems on copyrighted books has been running for years. This weekend it became considerably more concrete. The Authors Guild and its co-plaintiffs have asked a US federal court for summary judgment against OpenAI and Microsoft, arguing that the companies infringed authors’ copyrights by obtaining books from pirate sources, copying them for AI training and transferring copies between themselves.
The filing, announced by the Authors Guild on 5 September, asks the court to reject the defendants’ fair-use defence. No final ruling has been made, and OpenAI and Microsoft dispute the authors’ infringement claims. But this is an important stage because the plaintiffs are now asking the judge to decide key legal questions without waiting for a full trial.
For writers, the timing matters. Only days earlier, the US government urged G20 countries to create frameworks that allow AI developers to train on creators’ work under fair-use principles while also protecting creators. In other words, two competing visions are now sitting almost side by side: one treats access to creative work as important infrastructure for AI innovation; the other asks where innovation ends and uncompensated copying begins.
Training is not the only question
It is tempting to reduce the whole dispute to a single sentence: “Is AI training fair use?” That is too simple. The Authors Guild’s case puts particular emphasis on how books were acquired, whether copies came from pirate libraries, what copies were made during the process and how those copies were shared. Those questions may prove just as important as the abstract question of whether machine learning itself can ever qualify as fair use.
Reuters reported on 2 September that the US government urged a comparatively permissive international approach to AI training. That makes the legal distinction especially significant. A future in which some training is lawful does not automatically mean every route used to obtain a book, every intermediate copy, or every commercial use of that copy becomes lawful too.
Why writers should care even if they use AI themselves
This is also where the AI debate can become needlessly tribal. An author can oppose unauthorised use of books for model training while still using artificial intelligence as a practical tool. Those positions are not contradictory. The important distinction is what the technology is doing and who retains creative responsibility.
For my own work, the fiction remains mine. I create the characters, worlds, story, language and final manuscript. AI can sit around that process: helping visualise a starship or character, organising research, supporting advertising, preparing social material or taking some of the repetitive administration out of running an author business. That is very different from asking a machine to generate a novel and then presenting the result as though a human wrote it.
Publishing badly needs to preserve that distinction. If every use of AI is treated as equivalent, we end up punishing transparent authors who use modern tools while failing to deal clearly with genuinely machine-authored fiction. The useful test is not simply “Was AI involved?” It is: what did the AI do, what did the human create, and who is taking responsibility for the finished work?
The bigger issue is bargaining power
There is another reason this case matters. Individual writers have almost no practical bargaining power when dealing with the companies building frontier AI systems. Most authors cannot negotiate a training licence with a technology company, inspect a model’s dataset or discover easily whether their books were copied. That is why transparency, licensing mechanisms and collective representation are becoming as important as courtroom definitions of fair use.
The Anthropic copyright settlement has already shown that disputes over training data can translate into real claims and potential payments for authors. The Authors Guild updated writers on that settlement on 4 September as claim reconciliation continues. Different cases involve different facts, so one result will not automatically decide another. But together they are slowly building the legal and commercial architecture around AI and books.
What should authors watch next?
The immediate thing to watch is how the court handles the separation between model training, acquisition of source material and copying. A broad ruling that treats all of those activities as one undifferentiated act would have very different consequences from a decision that says training may sometimes be fair use but pirated acquisition or other copying can still infringe copyright.
Longer term, writers should watch for workable licensing systems and meaningful disclosure. I would rather see an AI economy in which authors can make informed choices about their work than one built around permanent litigation between creators and technology companies. Innovation matters. So does the principle that a creative work does not cease to have an owner merely because it has become useful to a machine.
Today’s question
If AI companies are ultimately allowed to train on copyrighted books under fair use, what should authors receive in return: transparency, payment, a genuine opt-out, stronger protection against pirated source material. Or some combination of all four?
Rob Frankson is a science-fiction author and creator of the Near Galaxy Saga. Through 121 Minutes he writes about storytelling, publishing, creativity and the changing relationship between authors and artificial intelligence.
AI & Editorial Transparency
AI & The Author is edited and published by Rob Frankson. Artificial intelligence is used to assist with news research, initial drafting, content organisation and supporting imagery. All articles are reviewed and, where necessary, edited by Rob Frankson before publication. The opinions, editorial position and final decision to publish remain the author's.
Comments