Amazon now has nailed its publishing industry AI colours to the mast. We cannot have one rule for Amazon and another for every other AI company.
Contrast these statements on AI:
“We will be hand delivering a letter to Meta detailing our strong objections to their current use of unlicensed works to train AI, demanding that authors are fairly remunerated and that their consent is sought for all future use” – Anna Ganley, CEO, the Society of Authors.
“If tech companies want to use creators’ copyright-protected work, they need to pay for that use and creators shouldn’t have to opt out from having their work stolen for the commercial benefit of AI tech companies. The UK has world-leading copyright laws which protect and promote our booming creative economy and enable creators to make a living. These laws apply to everyone – even big tech – and must not be eroded” – Anna Ganley, CEO, the Society of Authors.
On the UK government not acting on AI and copyright: “AI tools have benefited tremendously from the hard work of authors across the UK, without their knowledge, and against their will, and the government seems content to let these practices continue for the sake of protecting the prospects and interests of the tech sector” – Anna Ganley, CEO, the Society of Authors.
“The great copyright heist cannot go unchallenged. Big Tech needs to pay for the creative and research content they hoover up to train AI” – Dan Conway, CEO the Publishers Association.
The “teenage demands of a tech sector who want content for free” – Dan Conway, CEO the Publishers Association.
Go for it, Dan! Say it out loud! AI is stealing jobs from humans! It is stealing our IP and then stealing our careers!
And that’s why the PA is busily holding workshops for publishers to exploit AI. On 29 April the PA hosted the first event of the Publishers Association Insight Series focusing on how publishers can Integrate AI into business processes and Streamline and content management.
Great to see! But to be clear here, Dan Conway, CEO of the PA, is encouraging publishing professions to use the very AI he is claiming is built on theft. And of course the PA is on Instagram, YouTube and X, all tech companies that are using AI developed, according to the PA, in “the great copyright heist.”
No Joined-Up Thinking
Needless to say, the Society of Authors has organised a full boycott of X and Meta-owned Facebook and Instagram in protest at Meta’s actions. No, hold on, that would require joined-up thinking. Instead, the SoA put on a noisy show outside Meta UK’s offices, and then sent them a letter. By post. I’ve heard Mark Zuckerberg is quaking in his boots.
This past week, Amazon-owned Audible announced roll-out of AI for narration and translation of audiobooks.
“Audible believes that AI represents a momentous opportunity to expand the availability of audiobooks with the vision of offering customers every book in every language,” Audible CEO Bob Carrigan stated with not a hint of embarrassment. After all, those exact words – “every book in every language” – was what Jeff Bezos promised us with the Kindle store every year for its first seven years or so, before that statement was quietly dropped from the website.
Now, per the press release, “Audible’s integrated AI narration technology combines our nearly thirty years of experience producing exceptional audio stories with Amazon’s advanced AI capabilities. In the coming months, interested publishing partners will be able to choose one of two pathways to production:
- Audible-managed, end-to-end production: Through our end-to-end service, Audible handles the entire audiobook production process for selected titles, managing every step from initial text ingestion to published audiobook.
- Self-service production: Publishers access the same underlying technology while directing their own production process, independently producing high-quality audiobooks.
Across both pathways, publishers can choose from a quickly growing and improving selection of more than 100 AI-generated voices across English, Spanish, French, and Italian with multiple accent and dialect options, and will be able to access voice upgrades for their titles as our technology evolves.“
Oh yeah, and let’s not forget translations: “As we expand our global audience for audiobooks, we’ll also begin rolling out AI translation in beta later in 2025, allowing select publishers to bring their audiobooks to international audiences in their local languages. We’re developing support for translations from English to Spanish, French, Italian and German, which we’ll begin to roll out throughout this year. Publishers can opt for human review from professional linguists to ensure translation accuracy and cultural nuance, and will be able to review the translations themselves in our text editor.”
The Deafening Silence
Just to be clear there, human review will be an option, at extra cost, of course, but the default offering is machine-made without even human oversight. No wonder the industry’s watchdogs were up in arms.
Dan Conway, over at the Publishers Association, was quick to say nothing about this flagrant disregard for the jobs of narrators and translators using AI trained on stolen content.
While at the SoA, Anna Ganley really let rip, whispering in no uncertain terms that although this “will expand the availability of audiobooks and help writers of all profiles reach new audiences”, the opportunities offered “must be transparent both to authors and consumers. Audible’s AI tool, and others like it, must not be used as a back door to teach and refine existing AI tools. Authors must be included in the process and not get shut out by tech companies and publishers. They must also be able to choose whether their work is narrated by a human or synthetic voice, and this must be clearly labelled to consumers.“
And I’m sure Anna will even now be scribbling those thoughts into a nice hand-written letter to send by pigeon-post to Bob and Andy.
So how exactly do Conway and Ganley think Amazon has managed to develop AI tools good enough to offer this service to narrate and translate these audiobooks? By magic?
Amazon Not Mentioned
Back last August the SoA, convinced the AI companies had committed a heinous crime, but lacking any concrete evidence, wrote a letter to OpenAI, Google, Microsoft and Apple: “On behalf of our members we request that you identify the works which have been used to date to develop your AI model.“
Yes, you read right. The SoA wrote to the tech companies asking them to incriminate themselves.
The companies were given seven days to acknowledge and twenty-one days to confess to everything. Needless to say, the AI companies were too busy to reply and in September the SoA reported it had sent follow-up letters. Then, silence.
There is no suggestion that the SoA wrote to either Amazon or Anthropic on this topic. Which is curious given Amazon invested $4 billion in Anthropic as part of a broader partnership, and makes no secret of its plans for AI.
But it seems that OpenAI and Meta and co. only need to fart and the PA and the SoA will be there with their soundbite megaphones shouting “heist!“, “thieves!“, “teenage demands for free content” and “you didn’t ask permission and you must compensate us!”
Yet while OpenAI and Meta et al face lawsuits and protests over AI’s use of copyrighted works – Anna Ganley even sent Meta a letter by snail-mail, such was the SoA’s outrage – Amazon has quietly become a dominant force in AI-driven publishing innovation.
Audible Goes Big on AI Narration and Translation
Amazon’s AI initiatives have been coming thick and fast, some involving an intimate knowledge of book content (Recaps, anyone?, but no initiative has been bigger and potentially more worthy of outraged soundbites (if that’s what rocks your boat) than the May 13 announcement that Amazon-owned Audible was going big on AI narration and AI translation.
A reminder of Audible’s own words: “Bringing new audiobooks to life through our own fully integrated, end-to-end AI production technology.”
So of course the PA and SoA just looked the other way. One rule for Amazon, another for everyone else.
At which point, full disclosure. Issues of permission and compensation aside, I don’t have a problem with AI. Regular readers will know I believe AI will transform publishing in ways we are just beginning to glimpse right now. And Amazon has already transformed publishing. For all its faults, Amazon is an essential component of the modern publishing industry. I don’t have a problem with Amazon.
But in Amazon we have a unique situation, because Amazon actually has our music, our comics, our books, and our audiobooks, right there on its platform. With our permission.
Questions Galore
Did Amazon ask for permission to use that content for AI purposes? Clearly not. Is it compensating us for using that content for AI purposes, if that is happening? Absolutely not.
But clearly it trained its AI systems on something, and partner Anthropic is not noted for its licensing deals with book publishers.
Which begs the question why our protectors, who have so much spleen to vent about Meta and OpenAI, have so little to say about what Amazon is doing. Or do they seriously believe that Meta’s and OpenAI’s LLMs will steal our jobs, but Amazon AI somehow won’t? Do they seriously believe that OpenAI and Meta and co. are obligated to pay compensation and Amazon is not?
Is Amazon leveraging its platform strength, opaque contracts, and strategic partnerships to exploit creator content for AI training, while facing minimal backlash?
Just how much is Amazon using Anthropic, and how much is Anthropic using Amazon?
Are we looking at a two-tier system: negotiated protections for major publishers, and contractual traps for independents? Is Amazon hiding behind Anthropic’s nice-guy image to do its evil deeds?
And just what does the gobbledygook in an Amazon contract really mean?
The Amazon-Anthropic Partnership: Ethical AI or a Smokescreen?
In 2023, Amazon invested $4 billion in Anthropic, the AI startup behind Claude. Unlike rivals, Anthropic emphasises “ethical” AI training via:
- Licensed data: It claims to avoid scraped content, using only “publicly available or licensed works”.
- Opt-out tools: It allows rights holders to exclude content from training datasets.
Source: Anthropic’s Data Transparency Report (2024)
Critically, the report does not explicitly name Amazon, KDP, or Audible as data sources. Instead, it refers vaguely to “licensed partnerships,” which may or may not include content on the Amazon platform.
The Best Among The Rest
Before moving on here, let me just add that the consensus of opinion these past weeks is that Claude is very much at the top of the AI game. As April ended, Claude was acclaimed the best performing LLM out there. And I have to say its free tier warrants that accolade.
This is important, because pretty much every AI CEO has said, repeatedly, that access to quality copyrighted material is essential to maintain the highest standards.
So either Anthropic’s Claude has found some magic potion that no other AI company knows about, or Claude is being trained on copyrighted material from largely unidentified sources.
In 2023 Anthropic was sued by Universal Music Group and others regarding song lyrics allegedly regurgitated word for word. That case is still before the court, but it should be noted Anthropic has since introduced new measures to avoid similar transgressions.
In 2024 Anthropic was sued again, this time by book authors that claimed Anthropic used a dataset called “The Pile,” which incorporates nearly 200,000 books from a shadow library site, to train Claude. Per the Hollywood Reporter, this has been confirmed by Anthropic.
Accepting that Anthropic does not discuss its deals in detail, even so it is almost impossible to believe the company has somehow secured secret deals with major publishing houses, but the 200,000 books in “The Pile” is small beer in terms of AI content needs.
Might Anthropic’s deal with Amazon possibly includes access to content on the Amazon platform for training and like purposes? Or might the deal have allowed Amazon to train on the 200,000 titles in “The Pile”?
I’m not a lawyer so am not going to attempt to cherry-pick elements of the KDP and ACX contracts, but given Amazon’s $4 billion stake in Anthropic, it is not unreasonable to raise the possibility that:
- Amazon uses its licensed KDP/Audible content to train internal AI tools.
- Anthropic’s models may directly or indirectly benefit from this data via Amazon’s AI infrastructure (e.g., AWS AI services shared with Anthropic).
The bottom line is, we simply don’t know, and I must stress that Amazon’s keeping quiet about how it operates is standard policy, not something peculiar to the issue of AI, so we can read nothing into the apparent secrecy.
The Ethical Dilemma
But even if Anthropic doesn’t directly ingest KDP/Audible books, Amazon’s AI tools (trained on creator content) still power Anthropic’s infrastructure via AWS. This creates a “closed loop”:
- Authors upload books to KDP.
- Amazon trains AI tools on those books.
- AWS sells AI infrastructure to Anthropic.
- Anthropic’s models indirectly benefit from Amazon’s AI advancements.
Creators are excluded from this value chain.
Which brings us to compensation, a point so important to the PA and the SoA when any other tech company is in the firing line.
Be Absolutely Clear that Amazon is an AI Tech Company
Publishers get paid for sales/downloads, but get nothing for any other use of their uploaded content, which may or may not involve use to train AI systems.
But whether Amazon uses uploaded content, uses Anthropic’s content, or has another, as yet unidentified content source, we can be absolutely clear that Amazon is an AI company.
From the Audible press release:
“Audible’s integrated AI narration technology combines our nearly thirty years of experience producing exceptional audio stories with Amazon’s advanced AI capabilities”
“Publishers can choose from a quickly growing and improving selection of more than 100 AI-generated voices across English, Spanish, French, and Italian with multiple accent and dialect options, and will be able to access voice upgrades for their titles as our technology evolves.“
“As we expand our global audience for audiobooks, we’ll also begin rolling out AI translation in beta later in 2025.”
Yet while Conway and Ganley scream about how AI companies must pay their way, are thieves and vagabonds and want everything for free, Amazon somehow escapes their notice.
Well, almost. Earlier this month the SoA ran a piece about AI-generated books on Amazon, before of course Audible announced it would be selling its own AI created audiobooks). As part of that article, the SoA explained it had “asked Amazon about AI-generated books. We are grateful to them for their responses.”
So, Anna Ganley, how about another letter asking Amazon which of your members’ books it trained its AI system on, and how much it is paying in compensation? Or is it one rule for Amazon and another for everyone else?
Reasons for the Disparity in Outcry
Messaging and Framing: Amazon often frames its AI initiatives in terms of increasing accessibility, efficiency, and offering new options (e.g., making more books available as audiobooks). This can sometimes soften the immediate perception of job displacement or copyright infringement compared to announcements that are solely focused on automation or content generation without clear benefits for creators or consumers.
Scale and Integration: Amazon’s sheer scale and the deep integration of AI across its existing services might make it harder to isolate specific instances of AI use for criticism. AI is woven into the fabric of its operations, rather than being presented as a single, disruptive new product.
Market Dominance: As a dominant player in e-commerce, cloud computing (AWS), and digital content distribution (Kindle, Audible), Amazon holds significant market power. This position might influence how criticism is voiced or perceived within the industry, as I explored (with less focus on AI) a few weeks ago.
But with specific regard to AI:
Focus of Early AI Controversies: Much of the initial public and media attention on AI controversies has focused on generative AI models producing novel text, images, or code in a way that directly competes with human creators, often without clear attribution or compensation. While Amazon is moving into this space with AI narration, other companies might have been earlier or more prominent, not to mention easier targets for this type of criticism.
Complex Legal Landscape: The legal framework around AI training data and copyright is still evolving. While lawsuits are being filed against various AI companies, the outcomes are uncertain, and this legal ambiguity can influence the nature and intensity of public debate.
Mutual Benefits?
As we’ve seen, Amazon’s AI developments have so far generated little resistance. One reason may be how Amazon positions its AI: as an internal enhancement of its retail services or publishing platform, rather than a standalone consumer app.
For example, Amazon can say these new features serve its own data (customer reviews, Amazon’s book catalogue), presenting its AI as a partnership with content owners, not as some opaque data grab.
But the bottom line is, if Amazon is using uploaded content to train its AI models, it has not expressly sought the permission of creatives, and is not offering any compensation. And the PA and the SoA look the other way.
If the benefits of the platform are to be deemed compensation enough, then why cannot Meta argue the benefits authors receive from Facebook, WhatsApp and Instagram?
In Meta’s case, a court will likely decide the matter, if Donald Trump doesn’t pre-empt the show with an Executive Order.
Amazon has previously been flagged for internal copyright violations, where senior management instructed an employee to “ignore” breaches that should have been reported to the Amazon legal team. The rationale given was that other AI companies were doing it and Amazon was falling behind in the AI race.
As above, Amazon’s AI partnership with Anthropic is with a company being sued by both music and book industry stakeholder for copyright infringements.
None of the above is proof of Amazon’s wrong-doing. But the presumption of innocence is not being afforded to any other AI company.
Our industry needs to acknowledge that
a) Amazon now has nailed its publishing industry AI colours to the mast
b) we don’t know how its AI systems were trained and we need to ask ourselves why this is not a subject of intense scrutiny, and
c) we cannot have one rule for Amazon and another for every other AI company.
This post first appeared in the TNPS LinkedIn newsletter
