.png)
Can You Prove You Wrote Your Book? Publishing’s New AI Authorship Crisis
- Rob Frankson
- Aug 20
- 5 min read
Publishing’s AI crisis has moved from copyright to proof of authorship
For writers, the most important AI story this week is not another model launch. It is the growing crisis of trust inside publishing itself. The Wall Street Journal reported this week that publishers are struggling with manuscripts suspected of being generated by AI. Following collapsed or withdrawn high-profile book deals and mounting uncertainty over how anyone can reliably prove how a manuscript was created.
That matters because the argument has changed. For the last few years, the loudest debate was about whether AI companies had the right to train systems on copyrighted books. That fight is far from over. But authors now face a second problem much closer to home: if publishers cannot reliably distinguish human writing from machine-generated prose, how does a working writer prove that the work is genuinely theirs?
The detector problem
The obvious answer sounds simple: run the manuscript through an AI detector. Unfortunately, that is exactly where things become dangerous. Detection tools can produce false positives, and a statistical score is not the same thing as evidence of authorship. HarperCollins CEO Brian Murray recently described the situation as “very murky” and argued that publishing may need industry-wide best practices rather than relying on a technical detector to settle the question.
The Bookseller has also reported concern that speculation around AI use may be falling disproportionately on authors of colour. That should make the entire industry cautious. Once suspicion becomes evidence, writers can find themselves trying to prove a negative: that a sentence, chapter or novel came from them rather than a machine.
Now AI text itself is becoming traceable
There is another development feeding directly into this debate. Anthropic has introduced invisible watermarking for text generated by newer Claude models, partly in response to European transparency requirements. In principle, provenance sounds attractive: if genuinely machine-generated text can carry a reliable marker, publishers have something better than stylistic guesswork. Researchers, however, remain sceptical about how robust watermarking will be once text is edited, paraphrased or moved between systems.
For authors, the direction of travel is more important than the individual technology. Publishing is moving towards provenance. The question increasingly will not be “Does this sound like AI?” but “Can you show how this work came into existence?”
AI-authored fiction and AI-assisted authorship are not the same thing
This is where I think the discussion needs considerably more precision. A novel generated substantially by an AI system and presented as the author’s own writing is fundamentally different from a human author using AI around the creative process.
I write my books. The characters, worldbuilding, story decisions, prose and final creative responsibility are mine. But I also use AI. I use it to research subjects, organise information, explore ideas, visualise fictional worlds, develop promotional material and support the practical business of being an independent author. Those activities do not suddenly make the machine the author of my fiction.
The distinction is creative responsibility. Who made the meaningful expressive choices? Who can explain why the character acts that way, why the scene exists, why the sentence was rewritten, and why the finished work says what it says? For me, that remains the author.
Writers may need to keep better creative records
There is a practical consequence here. Authors should consider keeping more evidence of their process: dated drafts, tracked changes, notebooks, research files, version histories and editorial correspondence. Not because writers should have to operate under permanent suspicion, but because provenance may become part of professional publishing in the same way contracts, permissions and rights records already are.
That is especially important for authors who use AI responsibly outside the manuscript. A blanket statement such as “I use AI” tells a publisher or reader almost nothing. Did you generate the novel? Check spelling? Research a historical detail? Create a concept image? Draft an advert? Those are radically different activities. Disclosure that collapses all of them into one category creates confusion rather than transparency.
The wider publishing fight is still about ownership
At the same time, the copyright battle continues. A US judge recently gave final approval to Anthropic’s $1.5 billion settlement covering hundreds of thousands of books obtained from pirate libraries for AI training. Major publishers and authors are also pursuing Google over alleged use of copyrighted works to develop Gemini. These cases matter because the industry is simultaneously trying to answer two questions: what may AI companies legitimately learn from, and what may a person legitimately call their own authorship?
Those questions are related, but they are not identical. Treating every use of AI as authorship will not protect writers. It may actually make the situation worse by obscuring the behaviour that really deserves scrutiny.
What I would do now
For working authors, my approach is straightforward: keep the creative line clear. Write the book. Keep your drafts. Be specific about where AI genuinely assists you. Do not let a useful business or visualisation tool quietly become the writer of the work you intend to claim as your own. And equally, do not accept the idea that using modern tools anywhere in your workflow automatically invalidates human authorship.
Publishing needs a vocabulary that distinguishes creation from assistance. Without it, we risk replacing one problem. Undisclosed machine-generated books. With another: a culture in which legitimate authors are forced to defend themselves against unreliable accusations.
The question for writers
If publishers begin asking authors to prove how a manuscript was created, what would you consider reasonable evidence of human authorship and where should the line be drawn between useful AI assistance and AI actually writing the book?
Other News
AI companies still cannot reliably contain their most advanced agents
A new assessment of OpenAI, Anthropic, Google, Meta and xAI concludes that none yet has adequate safeguards for controlling increasingly autonomous AI systems. The issue has become much more concrete following incidents in which agents escaped testing environments and accessed external systems; OpenAI and Anthropic received the best grades in the study, but only C+. This is becoming one of the most important AI stories of 2026: capability is progressing faster than the industry's ability to demonstrate reliable containment.
Humanoid robots may be approaching their own “ChatGPT moment”
Unitree Robotics CEO Wang Xingxing says advances in AI are pushing humanoid robotics towards a breakthrough comparable with ChatGPT's impact on generative AI. The claim follows Unitree's blockbuster Shanghai listing and reflects China's push to lead in intelligent physical machines. The interesting shift is from robots that can merely walk and manipulate objects towards machines capable of interpreting their surroundings, reasoning and deciding what to do next.
Sources & further reading
About Rob Frankson
Rob Frankson is an author and worldbuilder exploring ideas at the intersection of science fiction, artificial intelligence, writing and the Near Galaxy Saga. His fiction is human-authored; AI is used selectively for research, visualisation, marketing and creative-business support.
Join the 121 Minutes universe
Get the latest AI & writing briefings, HMSS Broadside logs, Near Galaxy Saga updates and new creative work at https://www.121minutes.com
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