AI Copyright’s Transparency Problem: If Writers Must Disclose AI, Should AI Disclose Its Sources?

AI & The Author | 18 September 2026

Quick Shortcuts: The new court filing | Why authors should care | AI-assisted is not AI-authored | The question publishing cannot avoid

The AI Copyright Argument Just Became More Uncomfortable

For writers, the argument over artificial intelligence and copyright has often felt frustratingly abstract: training data, fair use, model weights, licensing and legal definitions that seem a long way from the desk where somebody is trying to finish a novel.

This week it became rather more concrete. Newly unsealed material in The New York Times copyright litigation against OpenAI and Microsoft has put internal discussions about publishers, paywalled material and the economic consequences of generative AI into public view. The Financial Times and Wall Street Journal reported on 17 September that employees had discussed the threat AI products could pose to publishers even as the companies developed systems trained on large quantities of online material. OpenAI and Microsoft dispute the plaintiffs’ copyright claims and continue to argue that AI training can qualify as fair use.

That distinction matters. An allegation in a court filing is not a judgement, and internal discussions are not themselves proof that copyright law was broken. But for authors the documents sharpen a question that has been sitting underneath this debate from the beginning: if an industry understands that the material created by writers and publishers has economic value, what responsibility does it have when using that material to build a competing commercial technology?

This is bigger than whether AI can write

Much of the public conversation still concentrates on whether AI can produce a convincing novel, screenplay or article. I think that can distract us from the more immediate issue. AI does not have to replace every writer to alter the economics of writing. It only has to change how creative work is discovered, summarised, reproduced, licensed and valued.

That is why the provenance of training material matters. Earlier copyright cases have already started separating two questions that are too often bundled together: whether training a model can be transformative, and whether the copies used to carry out that training were lawfully obtained. Authors should pay attention to both.

The current New York litigation is particularly important because newspapers and book authors are confronting many of the same underlying problem: their work is valuable precisely because humans invested time, expertise and money in creating it. Once that work becomes data, it is easy for the language surrounding it to make the original act of creation disappear.

AI-assisted is not AI-authored

There is another distinction worth defending just as strongly. Using artificial intelligence does not automatically mean handing authorship to a machine.

I write my books. I decide what the characters want, what happens to them, what the world means and which words finally stay on the page. AI can be useful around that work: research support, visualisation, organising ideas, exploring marketing, creating promotional material and handling parts of the increasingly complicated business of being an independent author.

Those activities are not the same as asking a model to manufacture the novel and putting my name on the result. Treating every use of AI as identical actually makes sensible rules harder to create. We need language capable of distinguishing assistance from substitution, just as we need rules capable of distinguishing licensed creative material from material simply taken because it was technically accessible.

Trust may become the real publishing currency

Writers are already being asked to think about disclosure, AI detection and proof of human authorship. Jane Friedman noted this month that there is still no universally accepted method or score that proves a manuscript is human-authored, even while agents and editors increasingly encounter submissions involving AI. That makes trust more important, not less.

Publishers, authors and AI companies therefore have a shared interest in clearer provenance. Writers should be able to explain how technology participated in their work. AI companies should be able to explain the basis on which creative work participated in their technology. Neither side benefits in the long term from ambiguity becoming the business model.

The opportunity is still there

None of this requires authors to reject AI. I remain interested in what these tools can do for a working writer, particularly the jobs surrounding the creative act that consume time without necessarily improving the story. Used deliberately, AI can give some of that time back.

But useful technology does not require us to abandon the principle that creative work has an owner, a history and a value. In fact, if AI becomes a permanent part of publishing, those principles become more important.

The industry now has an opportunity to move beyond the crude choice between banning AI and taking everything. Licensing, transparent data provenance, meaningful disclosure and clearly protected human authorship offer a more sustainable middle ground.

If AI companies expect writers to be transparent about how AI was used in our books, should writers expect the same transparency about how our books were used to build AI?

About Rob Frankson

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.

Want more?

Join 121 Minutes

For more writing, publishing, science fiction and AI & The Author, visit 121minutes.com and Join the 121 Minutes community for the latest AI & The Author briefings, writing updates and new fiction.

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.