.png)
The US Backs AI Fair Use: What It Could Mean for Authors and Their Books
- Rob Frankson
- 5 hours ago
- 4 min read
AI & The Author — 3 September 2026
The US government has stepped directly into one of the most important copyright fights of the AI era. In a court filing on 2 September, the Justice Department backed OpenAI in its dispute with The New York Times, arguing that training large language models on copyrighted material can qualify as fair use. On the same day, US Commerce Secretary Howard Lutnick urged G20 countries to develop frameworks that allow AI companies to train on creators’ work while balancing protection for rights holders.
For writers, this is bigger than one lawsuit. The argument over whether books, journalism and other creative work can be copied into training systems without individual permission is moving from courtrooms into national industrial policy. The question is no longer simply what AI can do. It is who gets to decide what happens to the human work that makes those systems useful.
Quick Shortcuts
The fair-use argument just became government policy
The Justice Department’s intervention matters because it gives the fair-use argument the weight of the US government. Its position is that AI training is transformative and important to scientific progress, economic development and national security. The New York Times and other rights holders take a very different view: that companies should not be able to ingest valuable copyrighted work at industrial scale and then build competing commercial products without permission or compensation.
That disagreement has existed for years. What changed this week is the political level at which it is now being argued. At the G20 technology meeting, Lutnick urged other countries to adopt frameworks favourable to AI training. If that approach gains international support, the rules governing authors’ work may be shaped as much by competition between nations as by traditional copyright principles.
Why writers should care
For an author, copyright is not an abstract legal mechanism. It is the foundation of the bargain: I create the work, I own certain rights in it, and I decide how those rights are licensed. AI complicates that bargain because training is largely invisible to the creator. A book can be copied, broken into data and incorporated into a model without the reader-like transaction we normally associate with a published work.
There is also an important distinction between training and output. Courts may ultimately decide that some forms of model training are fair use while still finding that pirated acquisition, verbatim reproduction or competing outputs cross legal lines. The recent Sony and Warner action against Anthropic underlines that point: the publishers allege unlawful acquisition and reproduction of protected material, while Anthropic says it will defend its practices.
AI-authored is not the same as AI-assisted
This debate also needs more precision inside the writing community. A novel generated substantially by a machine and presented as human-authored raises obvious questions about authorship, originality and disclosure. That is not the same activity as a working writer using AI to visualise a character, organise research, test marketing ideas, create supporting imagery or help manage the business surrounding independently written fiction.
That distinction matters to me. I write my fiction. The characters, worlds, decisions and final words are mine. I am also willing to use AI as a tool around that creative work where it is useful and where I can be transparent about what it has done. Treating every interaction with AI as equivalent to handing over authorship makes the discussion less useful, not more.
The uncomfortable double standard
There is a growing expectation that authors should be able to explain exactly how AI touched their work. That is reasonable when disclosure affects readers, publishers or collaborators. But the same standard of provenance should run in both directions. If creators are expected to account for the tools they use, AI developers should be able to account for the material their systems were built from.
That does not automatically mean every training use requires the same licence or payment. Copyright law has always contained exceptions and balances. But “innovation” should not become a substitute for answering basic questions about where material came from, whether it was lawfully acquired and what happens when a commercial system reproduces or substitutes for the original.
What happens next
The New York Times case is therefore worth watching well beyond journalism. Its reasoning could influence how courts treat books and other written works, while the US push at the G20 shows that governments are already thinking about AI training as strategic infrastructure. Authors may find themselves caught between two powerful ideas: the right of creators to control valuable work and the desire of nations to give their AI industries access to enormous quantities of high-quality data.
A workable settlement probably needs more than a binary choice between banning AI training and declaring everything fair use. Better provenance, lawful acquisition, practical licensing systems, meaningful transparency and clear rules around reproduction could give both innovation and creative rights somewhere to stand.
The question for writers
If governments decide that AI companies should be allowed to train on copyrighted books under fair use, what protections should authors receive in return — transparency, payment, an opt-out, limits on reproduction, or something stronger?
About Rob Frankson
Stay Connected with 121 Minutes
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