AI rights contracts have moved from a side concern to a front-line issue in book deals, especially after contract changes reported at Penguin Random House and Simon & Schuster. For authors, the practical question is not whether a publisher uses technology somewhere in its workflow. The sharper question is what the contract allows, what it forbids, what the author must disclose, and whether any consent is specific enough to control future uses of the manuscript.
This is not legal advice. Authors should review proposed language with an agent, publishing attorney, or other qualified adviser before signing. Still, every author can learn to spot the pressure points: training rights, benchmarking, internal tool use, output rights, warranties about AI assistance, and the difference between permission that is express and permission that is implied.
Why AI Rights Contracts Became A Front-List Issue
Penguin Random House Set A Contract Marker
According to PublishingPost, Penguin Random House introduced its first standardized AI clause in October 2025 under a new subsidiary-rights subsection called “Computational and Derivative Licensing,” and agents described the language as strongly pro-publisher PublishingPost reporting. The placement matters. Subsidiary rights sections usually deal with rights beyond the core print or ebook publication grant, so putting AI language there can shape how training, derivative uses, or machine-assisted exploitation are treated.
The same reporting noted that contracts signed before January 2025 often lacked coherent AI clauses, while newer standard drafts increasingly contain AI terms. That split leaves authors with two tasks. For new deals, they need to negotiate before signing. For older books, they may need to read the original grant of rights carefully before assuming that any AI-related use is either allowed or barred.
Simon & Schuster Showed Why Small Phrases Matter
The research notes for this brief state that Simon & Schuster added a one-sentence opt-out provision to its standard author contracts in March 2026, and that agents flagged it as difficult to enforce. That is a useful warning for AI rights contracts: a sentence that appears protective may not give the author workable control unless it identifies the covered works, covered systems, publisher duties, consent process, recordkeeping, and consequences if the term is breached.
Research also indicates that Simon & Schuster inserted further AI-related language into some author contracts in late May 2026, including a warranty that work was not significantly generated using large language model systems and consent language tied to internal editorial and marketing tools. Authors should not treat those details as universal across every S&S agreement. The safer practice is to review the actual draft in front of you, not a summary of what another author may have received.
Consent Should Be Specific, Not Assumed
Training Rights Need Separate Treatment
One of the clearest negotiation points is whether a publisher may license, upload, process, or otherwise use the manuscript for AI training. A June 2024 Romance Writers Report article stated that the Authors Guild updated its model contract to include an AI clause, and that Penguin Random House, Hachette Book Group, and Macmillan had adopted boilerplate saying they would not license or use an author’s work to train AI without express permission June 2024 RWR article.
That phrase, “express permission,” is doing important work. It suggests a separate, affirmative approval rather than silence, boilerplate, or a broad grant that sweeps in future technologies. In negotiation, an author might ask that any AI training use require a written amendment identifying the work, the system, the purpose, the duration, whether compensation applies, and whether the approval can be withdrawn for future use.
Internal Tools Still Deserve Boundaries
Publishers may argue that internal editorial, marketing, or production tools are different from commercial training licenses. That may be true in some situations, but the contract should say so clearly. A manuscript uploaded to a third-party tool for copy suggestions, audience analysis, sales positioning, translation support, narration testing, or marketing copy generation may create different risk questions depending on the vendor terms, retention policy, security controls, and whether the text is used to improve the tool.
Authors can ask for language distinguishing ordinary internal workflow from training or benchmarking. They can also ask the publisher to avoid submitting full manuscripts to external AI systems unless the vendor contract bars training, limits retention, protects confidentiality, and permits deletion. For a deeper treatment of permission-based drafting, see this related discussion of AI consent clauses.
Warranty Language Can Shift Risk To The Author
Disclosure Is Different From A Ban
Many publishing AI clauses do not begin with training rights. They begin with warranties. A warranty may require the author to state that the manuscript is original, that it does not infringe third-party rights, or that AI tools were not used in a prohibited way. In AI rights contracts, that language can become risky if “AI assistance” is undefined.
Authors should ask whether the clause covers only generative drafting, or whether it also reaches spelling tools, grammar tools, transcription software, research organization tools, accessibility tools, translation aids, image references, or brainstorming. The contract should avoid forcing an author to make an absolute promise about minor or incidental technology use if the publisher’s true concern is significant machine-generated text presented as the author’s original expression.
Records Help Both Sides
A practical compromise is a disclosure schedule. Instead of vague assurances, the author can identify any significant AI-assisted material, describe the kind of assistance, and confirm that the author reviewed and controlled the final text. The publisher can then decide whether further review is needed before acceptance.
This approach does not solve every issue, but it reduces surprise. It also gives editors, agents, and legal reviewers a written record if questions arise later. Authors should keep drafts, notes, tool disclosures, permissions, and correspondence organized. That habit is useful even outside AI, because publishing disputes often turn on what the contract and related records actually say.
Negotiating AI Rights Contracts With PRH And S&S Context
Ask For A Carve-Out Before The Deal Is Final
Research supplied for this article indicates that agents increasingly ask for AI exclusions and carve-outs, and that some Big Five imprints have accepted explicit exclusions rather than walking away from deals. Authors without high-profile bargaining power should not assume every request will be accepted. They can still ask precise questions early, before the contract is treated as settled.
A basic carve-out might say that no rights are granted to use the work, manuscript, excerpts, metadata, or derivative material for AI training, model improvement, benchmarking, synthetic narration, machine translation publication, or output generation without the author’s separate written consent. If the publisher insists on some internal use, the author can ask to limit it to non-training purposes tied to editing, production, marketing, or accessibility, with confidentiality and vendor restrictions.
Do Not Let Backlist Language Slip By
Backlist works deserve special care because older contracts may not speak clearly about AI. The research notes report that Simon & Schuster has been linked to a possible multi-year backlist licensing deal with a significant AI developer, involving roughly 200,000 titles, but also state that the deal was not confirmed as of August 2026. Because that detail was reported as unresolved, authors should treat it as a prompt to check their own contracts rather than as proof that any specific title has been licensed.
If an author has older books with a publisher, the useful step is to request written clarification. Ask whether the publisher believes it holds AI training rights, whether it has licensed the work for any AI-related use, whether the author can opt out, and whether the publisher will confirm that no such license will be granted without separate permission. Those questions should be asked in writing and answered in writing.
Clause Points Authors Can Raise Before Signing
AI Rights Contracts Requests Authors Can Put On The Table
Before signing, authors can prepare a focused list of contract requests. The list should be practical, tied to actual contract language, and short enough for an agent or attorney to work with efficiently.
- Define “AI system,” “training,” “benchmarking,” “model improvement,” and “AI-generated output.”
- Require separate written consent for training or licensing the work to an AI developer.
- Exclude manuscript uploads to third-party tools that retain text or use it to improve models.
- Limit any internal AI use to stated editorial, production, or marketing purposes.
- Clarify whether the author must disclose significant AI assistance and what counts as significant.
- Require notice if the publisher wants to create synthetic narration, machine translation, or derivative AI output.
- State whether payment, approval, attribution, and withdrawal rights apply to any later AI use.
The goal is not to block every ordinary technology tool used by a publisher. The goal is to prevent a broad rights grant from becoming permission for uses the author never knowingly approved. For authors comparing how specialized editorial sites handle rights, permissions, and niche audiences, the related network site Old Norse News offers a useful reminder that subject expertise and clear publication terms often belong together.
AI Rights Contracts For Author Control
AI rights contracts are now part of ordinary publishing due diligence. Penguin Random House’s standardized clause, Simon & Schuster’s reported contract moves, the Authors Guild model language, and agency pushback all point to the same lesson: authors should not wait until after signing to ask what AI permissions they have granted.
The strongest author position is built before the contract is finalized. Ask for definitions. Separate training rights from internal workflow. Treat warranties with care. Keep backlist questions in writing. Seek express-permission language where possible. If the publisher rejects a request, ask for narrower wording rather than accepting a vague clause that may be hard to police later.
No article can predict how a publisher will respond to a specific author, imprint, genre, or deal. The practical standard is simpler: if a manuscript may be used by people, vendors, or systems beyond the ordinary publication process, the contract should say what is allowed, what is not allowed, and who must approve the next step.