AI and Traditional Publishing: What's Actually in the Contracts
Reporting from the new frontier
Here's a legal gap that's been quietly causing problems in traditional publishing: most standard contracts were written before AI was a consideration, and the scramble to catch up has left room for some pretty broad interpretation.
The core issue: Older publishing contracts don't mention AI at all — which means they also don't prohibit it. Some publishers have leaned on existing broad-rights language — phrases like "internal purposes," "research," or "data mining" — and interpreted it as covering AI training too. The Authors Guild has flagged this directly, noting that such clauses "sometimes refer to AI expressly, but other times they are more oblique," and advising authors to watch for exactly that kind of vague wording.
It goes beyond training data. Two other issues have come up industry-wide:
Publishers using AI for audiobook narration or translation without requiring the author's explicit sign-off
Publishing staff uploading full manuscripts and authors' personal information into consumer AI tools (like ChatGPT) to generate summaries or marketing copy — without author permission, and without any guarantee the manuscript wouldn't be retained or used to train that tool
The second one stacks a privacy problem on top of a consent problem: it's not just "did I agree to this AI use" — it's "did my unpublished work just end up somewhere I have no visibility into."
The industry's response, so far: The Authors Guild has introduced model contract clauses that authors and agents can now request. The baseline clause states that publishers acquire no rights to use a book for AI training or AI-generated outputs unless those rights are expressly granted — reversing the old assumption that silence means permission. A second option lets authors selectively license specific AI uses for negotiated compensation. Additional clauses require the author's written consent before AI is used for audiobook narration, translation, or cover art.
The catch: none of this is automatic. These are opt-in requests, not industry-wide standards. Whether an author is protected still depends on whether they, or their agent, know to ask for these clauses — and whether the publisher agrees to include them.
Where it also gets contentious: the money. When a publisher does license a backlist for AI training, the Authors Guild's position is that authors should generally receive somewhere between 75–85% of the revenue, depending on the publisher's role in the deal. In practice, splits have varied a lot — some academic presses have reportedly offered authors as little as 25%.
The bottom line: This isn't a hypothetical or an overreaction — it's a real, current mismatch between how quickly AI adoption is moving on the publisher side and how slowly contract language is catching up to address it. The Authors Guild's clauses are a meaningful step toward closing that gap, but for now, protection is still something authors have to actively ask for rather than something they can assume they already have.
For further research:
Comments