Games Workshop has been one of the most vocal major companies in the gaming and hobby space when it comes to opposing artificial intelligence. Its websites carry explicit warnings against using their content for AI training. It banned internal use of AI tools in January 2026. It’s investigated its own marketing materials when fans suspected AI involvement. So in one sense, what happened at the end of the audiobook version of The Wicked and The Warped — Robert Rath’s Necrons novel from Black Library — shouldn’t be surprising. But hearing narrator Richard Reed actually read out an AI training prohibition out loud, in the middle of a recording, is still a genuinely striking moment.
The warning Reed reads at the end of the audiobook goes like this: “This recording is copyrighted to Games Workshop 2026. All rights reserved. May not be used or reproduced in any manner for text and datamining activities, or developing or training artificial intelligence models or systems.”
It appears to be the first time such a warning has been read aloud in a Black Library audiobook, and it lands at a moment when the entire question of AI companies training on books without permission has become one of the most contentious legal and ethical battlegrounds in the creative industries.
What Games Workshop Has Done on AI So Far
The audiobook warning doesn’t come out of nowhere. Games Workshop has built a fairly consistent anti-AI position over the past year or so, and its track record is worth laying out:
- January 2026 — Games Workshop formally banned staff from using AI in content production and design processes, with CEO Kevin Rountree stating that none of its senior managers were excited about the technology.
- Black Library website — The official home of Games Workshop’s book publishing carries a clear warning: “Any use of website content to train generative artificial intelligence (AI) technologies is expressly prohibited.” Similar language appears in the detailed legal sections across Games Workshop’s web presence.
- July 2026 — Rountree publicly reiterated the company’s anti-AI stance and confirmed that Games Workshop had launched internal investigations after some marketing images raised suspicions among fans that AI tools might have been used. The investigation concluded the incidents were “human errors” rather than AI usage, with Rountree stating: “They were investigated fully to ensure that AI had not been used and that all our staff are following our internal policies.”
The spoken AI warning in The Wicked and The Warped is an extension of this stance into the audio format — and it makes sense as a move, even if it’s an unusual one to actually hear. Audiobooks are recordings that get ingested by data scraping tools just like text can be, and the warning’s specific language about “text and datamining activities” alongside the AI training prohibition covers exactly that vector.
The Broader Context — AI Companies and Books
To understand why this feels meaningful right now, you need to understand where the AI-and-books legal battle stands in 2026. The situation is a mess of ongoing litigation, recent rulings, and investigative reporting that paints a complicated picture of what AI companies have been doing and what the courts have so far let them get away with.
The Anthropic Lawsuit and Settlement
The highest-profile case involving books and AI training concerned Anthropic, the company behind the Claude AI. A class-action lawsuit brought by authors revealed that Anthropic had run an internal initiative called Project Panama, through which the company purchased physical books in bulk — prioritising “less common” and hard-to-find titles — had their bindings removed with cutting equipment, and scanned the pages using industrial imaging equipment in what the documents described as “destructive scanning.” The stated internal goal was ambitious to the point of being alarming: Anthropic’s internal documents described plans to “destructively scan all the books in the world.”
US District Judge William Alsup issued a mixed ruling in June 2025. He found that Anthropic’s use of legally purchased books to train AI was fair use — the act of converting a physical book into a digital file for internal model training being “exceedingly transformative” in his view. However, Alsup found separately that Anthropic had also pirated more than 7 million books from shadow libraries including Books3, LibGen, and Pirate Library Mirror, and that this piracy was not protected by fair use. The case eventually settled — though the settlement terms were not publicly disclosed.
The fair use ruling on legitimately purchased books is the part that matters most for the creative industry’s concerns. Courts have so far been willing to let AI companies argue that training on books they legally own is transformative use — which provides very limited practical protection for authors and publishers who want their work excluded from training datasets.
Amazon and the Book-Scanning Reports
More recently, a report by 404 Media alleged that Amazon has been purchasing “massive quantities of books” — including rare physical copies — scanning them for AI training data, and destroying the originals in the process. The reporting described orders ranging from hundreds to potentially millions of books being processed in this way. Given what the Anthropic case revealed about the scale of physical book acquisition for AI training purposes, the Amazon allegations fit a broader pattern of AI companies treating the physical book market as a supply chain for training data.
The LibGen Connection and Warhammer Books
For Games Workshop and Black Library specifically, the most direct evidence of AI training on their content comes from the Meta LibGen revelations. An investigation by The Atlantic in March 2025 found that Meta allegedly pirated millions of books from LibGen — the pirated-books database — to train its AI models. Black Library authors were among those who subsequently discovered their novels were in the LibGen dataset. A search of the database shows virtually all of Robert Rath’s Warhammer novels — the same author who wrote The Wicked and The Warped — are present in that dataset.
This is the uncomfortable reality that Games Workshop’s warnings are responding to: the content protections on its websites and now in its audiobooks are legally defensible statements of intent, but they don’t prevent AI companies from using LibGen copies that were scraped without permission in the first place. The warning at the end of Reed’s narration doesn’t undo the fact that those books are likely already in training datasets.
The Fake Warhammer Books Problem
There’s a separate but related issue that illustrates what happens when AI models are trained on franchise-specific content: fake AI-generated Warhammer books have become a genuine problem for Black Library and its readers. Unlicensed Warhammer 40,000 novels — complete with obviously AI-generated cover art and AI-generated prose — have been appearing on retail platforms. They’re often indistinguishable from genuine Black Library releases to casual buyers, and they’re exactly the kind of output that becomes possible when AI models absorb enough of a franchise’s distinctive voice and lore.
This is the commercial harm end of the AI training debate — not just the abstract question of whether authors should be compensated for training data, but the concrete reality that AI trained on Black Library content can produce convincing-enough imitations to confuse buyers and dilute the market for legitimate releases.
Can Games Workshop Actually Stop This?
Here’s the honest answer to the question that the original article raises, and it’s not encouraging for Games Workshop: probably not, with the tools currently available to them.
The legal situation as it currently stands is that AI companies can argue fair use when training on legitimately acquired content, and courts have so far been receptive to that argument. Games Workshop’s website warnings and audiobook disclaimers are clear statements of intent and potentially useful in any future litigation, but they don’t have the force of law in a way that automatically prevents training. A company that wants to acquire and train on Black Library content can simply purchase the books and audiobooks, and the current legal landscape gives Games Workshop limited ability to stop them.
To truly test its position, Games Workshop would need to file a lawsuit against an AI company it believes has trained on its content — something it hasn’t done yet. Short of that, the warnings serve two purposes: they’re a legal paper trail for any future action, and they’re a public statement of values for a company that has built significant goodwill with its community around rejecting AI in creative production.
Rountree has acknowledged the limits of what Games Workshop can control, noting that AI is often “automatically included in third-party software whether we like it or not” — which makes it effectively impossible for the company to say definitively that no AI touches its pipeline at any point. “This is something we are just going to have to potentially live with and monitor,” he said.
What the Audiobook Warning Actually Achieves
Cynics will say that hearing Richard Reed read out an AI training prohibition at the end of a Necrons audiobook accomplishes nothing practical. And they’re not entirely wrong. The warning doesn’t create new legal protections that didn’t already exist. It doesn’t reach back and remove Black Library content from datasets it may already be in. It doesn’t stop anyone who’s determined to use the recording for training purposes.
What it does do is make the prohibition audible, personal, and impossible to miss. A legal notice buried in a website’s terms and conditions is something most people never read. A narrator speaking directly to the listener — the same voice they’ve been trusting to deliver hours of Warhammer fiction — stating clearly that this recording may not be used to train AI is a different kind of communication. It makes the issue real and immediate in a way that boilerplate text doesn’t.
Whether Games Workshop escalates from warnings to legal action — whether it eventually becomes the company that forces a definitive court ruling on AI training and creative IP — remains to be seen. For now, it’s carved out a clear position and is defending it as publicly and consistently as it can. The voice at the end of The Wicked and The Warped is a small but pointed part of that effort.
This week has had no shortage of stories about companies navigating the collision between new technology and existing rights. The tariff refund situation showed very different approaches from Panic and Nintendo when it comes to customer obligations. The Witcher 4 physical edition story highlighted how Sony’s disc end-date is creating problems for publishers who had no say in the decision. And the massive 12TB Valve data leak is still being worked through by the community — a reminder that data control is something even well-resourced companies struggle to maintain.
