How Risk Aversion is Reshaping AI Copyright Disputes – And What It Means for a Book Publishing Industry Still Searching for Leverage
The Warner Music Group-Suno settlement arrives with the inevitability of a pre-determined chess move.
Following Universal Music Group’s October settlements with both Udio and Suno, WMG’s announcement feels less like news, more like confirmation: the music industry has collectively decided that negotiated partnerships beat courtroom gambles. Yet this emerging pattern – where settlements supplant judgments and collaboration replaces confrontation – reveals starkly different dynamics between music and book publishing, with implications that my earlier analysis only partially anticipated.
From the post three weeks ago: “Technology evolves, business models adapt, and what seems impossible today may be standard tomorrow. The question is whether book publishers will help shape that evolution or simply respond to it once others have defined the landscape.”
“Technology evolves, business models adapt, and what seems impossible today may be standard tomorrow. The question is whether book publishers will help shape that evolution or simply respond to it once others have defined the landscape.”
That was the UMG settlement. This week, it’s WMG that has settled.
And here’s the thing: The WMG-Suno deal follows a now-familiar template. Financial terms remain undisclosed, but the partnership includes compensation for artists and songwriters, licensing commitments for new AI models, phasing out of existing unlicensed models, download restrictions for users, and artist opt-in requirements for name, image, likeness, and voice use.
Suno’s acquisition of Songkick from WMG as part of the deal suggests this isn’t merely litigation management but genuine strategic integration – the kind that makes lawyers on both sides sleep better at night (more on lawyers below).
What’s striking isn’t just that settlements are happening, but the speed at which they’re proliferating.
What’s striking isn’t just that settlements are happening, but the speed at which they’re proliferating. UMG and WMG have now both settled with Udio, and WMG becomes the first major label to partner with Suno despite Sony still pursuing litigation against both platforms. The gunfight at the AI Corral isn’t happening. Everyone’s checking their weapons at the door and sitting down for whiskey instead.
The Risk Calculation: Why x% of Something Beats 100% of Nothing
Okay, lawyer time. Lawyers aren’t as stupid as they look. They understand that that compromise is a win-win for both sides, while outright victory is guaranteed for none.
And this cuts to the heart of why this settlement cascade feels inevitable: for the plaintiffs’ lawyers in the Anthropic case, the class action structure meant they stood to lose everything with an adverse court decision, or gain a respectable payout from a mid-way settlement. This makes settlements increasingly attractive as precedent for future disputes – not because anyone’s proven their legal case, but because partial victories eliminate catastrophic downside risk.
Here to acknowledge, before looking at book publishing’s options, that the music industry’s position differs fundamentally from book publishers in this calculus.
Suno had just closed a $250 million funding round valuing the company at $2.45 billion, whilst Udio had raised only $10 million when it settled with UMG, with Barclays Research noting it could be “prohibitively expensive to lose”.
These AI music platforms have substantial capital and scaling user bases – they’re not judgment-proof startups that can settle for pennies. Moreover, they operate in a space where music labels control genuinely irreplaceable assets: the actual recorded performances and voices that fans want to interact with.
Book publishing’s position is weaker in three critical ways.
Book publishing’s position is weaker in three critical ways.
First, the training data question differs structurally. The judge in the Anthropic case ruled that training AI on legally obtained copyrighted books was transformative fair use, but that Anthropic’s use of pirated books from shadow libraries was infringement, with the settlement addressing only “the narrow claims about how certain materials were obtained”. This means the core issue – whether training on legally acquired books constitutes infringement – was decided in Anthropic’s favour before the settlement. The $1.5 billion payout compensated only for the method of acquisition, not the use itself.
Second, the economic leverage differs dramatically. Taylor Swift’s voice and Ariana Grande’s recordings can’t be legally replicated through fair use. But the underlying narratives, character types, plot structures, and even writing styles in books may be learnable from legally obtained copies without infringement. An AI trained on public domain literature, legally purchased books, and licensed text could potentially generate commercially viable content without ever touching pirated material. Music’s leverage comes from controlling irreplaceable performances. Book publishing’s leverage comes from… what, exactly? Copyright in specific expression, yes, but that’s the very question courts are deciding in publishers’ disfavour.
Third, book publishing is more fragmented than music, where UMG is the world’s biggest music label with artists including Taylor Swift, Ariana Grande et al. Would AI companies offer comparable partnership terms to Penguin Random House, let alone mid-sized publishers? Or would fragmented negotiating power create unfavourable arrangements where publishers accept terms they wouldn’t choose if they could coordinate industry-wide responses?
What My Earlier Analysis Got Right – and Wrong
Looking back at my early November analysis through the lens of these latest settlements, several predictions hold up whilst others require revision.
Where the Analysis Holds
The fundamental observation that one settlement emphatically does not set legal precedent in the formal sense, as settlements are negotiated agreements rather than court rulings remains true – but the commercial precedent is now undeniable. The pattern across UMG-Udio, UMG-Suno, and WMG-Udio/Suno establishes a template: major content holders are choosing partnership over protracted litigation. The domino effect has arrived.
The structural differences between music and book publishing I identified remain valid. The Udio platform operates as a “walled garden” where AI-created music stays within Udio and doesn’t compete on Spotify or other streaming services, creating a bifurcated market that books don’t have an equivalent consumption model for. Yet.
The WMG settlement reinforces this: only Suno’s paid tier subscribers will be able to download their creations off-platform, with paid users having download caps requiring payment for additional downloads. This directly addresses streaming service flooding whilst preserving platform monetisation. Books lack this kind of controlled distribution ecosystem.
The Kindle Worlds comparison I drew proves increasingly relevant. Amazon’s experiment enabled writers to create fan fiction based on licensed properties for royalties, with Amazon paying fan-authors 35% of net revenue for works over 10,000 words whilst paying royalties to original rights holders – remarkably similar to what UMG and Udio built. The failures I identified – community rejection of commercialisation, rights issues, and platform abandonment – remain cautionary. WMG CEO Robert Kyncl’s emphasis on artist opt-in for “name, image, likeness, voice and compositions” suggests music has learned these lessons about creator control.
Where Revision Is Needed
I underestimated how quickly music would converge on collaborative models. Writing in early November 2024, the UMG-Udio settlement seemed potentially anomalous. Weeks later, we have four major settlements across two AI platforms, with only Sony still litigating. The velocity suggests strategic coordination – not necessarily collusion, but certainly pattern recognition. Once UMG demonstrated that partnership worked, WMG faced a different calculation: let competitors define the terms, or negotiate from strength while leverage exists?
I also underestimated the immediacy of the platform economics. Suno’s valuation at $2.45 billion following its $250 million funding round demonstrates these aren’t experimental side projects like Kindle Worlds but serious commercial platforms with investor confidence. The funding came from Menlo Ventures, NVIDIA’s VC arm, and a music industry veteran – suggesting convergence between tech capital and music industry thinking.
The music settlements suggest collaboration is something AI platforms will increasingly demand as the price of doing business.
Most significantly, I framed collaboration with AI companies as something publishers might consider. The music settlements suggest it’s something AI platforms will increasingly demand as the price of doing business. Suno and Udio aren’t begging for licenses; they’re raising hundreds of millions in capital and integrating music partnerships into their core business models. They’re building sustainable platforms with or without full industry participation – which means holdouts risk marginalisation rather than extraction of maximum value.
The Book Publishing Dilemma: Leverage Without Precedent
This is where book publishing’s position becomes genuinely precarious. The music industry is settling from a position of strength: platforms need their irreplaceable content, they have billions in user engagement to monetise, and labels can credibly threaten litigation that might destroy business models.
Book publishers are settling from a position of weakness: courts have ruled training on legally obtained books constitutes fair use, AI platforms can potentially generate commercially viable content without publisher participation, and fragmented industry structure limits coordinated negotiating power.
The Anthropic settlement illustrates this asymmetry. Authors’ attorneys described it as “the largest publicly reported copyright recovery in history,” being four times the minimum statutory damages – yet it covers only past conduct through August 25, 2025, doesn’t cover future uses or claims based on infringing LLM outputs, and Anthropic must destroy the pirated datasets within 30 days. Crucially, the judge’s June ruling that legally obtained books used for training constitutes fair use remains intact, meaning the settlement doesn’t resolve the fundamental fair use question.
The music settlements aren’t just about money – they’re about ongoing partnership.
This matters enormously. The music settlements aren’t just about money – they’re about ongoing partnership. Warner Music described the deal as a “landmark pact” that “expands revenue and delivers new fan experiences”, with Suno CEO Mikey Shulman promising “new, more robust features for creation, opportunities to collaborate and interact with some of the most talented musicians in the world”. These are forward-looking business relationships, not merely compensation for past wrongs.
Book publishers’ settlements look different. The Anthropic deal paid handsomely for past infringement but established no ongoing partnership, no licensing framework, no collaborative platform development. It was a cheque for using pirated copies, full stop. Anthropic can now train future models on legally acquired books without publisher permission or compensation, relying on the fair use determination that survived the settlement.
Why Safe Settlement May Not Be Available to Publishers
The framing that safe settlement beats possible courtroom catastrophe assumes both parties perceive similar downside risk. For music, this symmetry exists: platforms risk injunctions that could destroy their business models, whilst labels risk precedent that training on copyrighted material is categorically fair use. Both sides have existential stakes, making negotiated middle ground rational.
For book publishing, the risk asymmetry tilts heavily toward AI companies. They’ve already won the key legal question: training on legally obtained books is fair use. Their remaining risk is statutory damages for pirated materials, which they can eliminate by switching to legal acquisition methods. Publishers’ remaining leverage is… the threat of litigation over acquisition methods? That’s a dramatically weaker position than music’s “you need our irreplaceable content to make your platform work.”
This suggests book publishers may face a stark choice: accept unfavourable partnership terms that generate modest revenue streams, or watch AI companies train on legally purchased books whilst publishers receive nothing beyond retail book sales. The music industry’s “safe settlement” option may not be available at equivalent value.
The Collaborative Models Revisited: What’s Actually Viable?
My earlier analysis explored several potential collaborative models for book publishing, drawing from both the Udio framework and Kindle Worlds lessons. The latest settlements provide evidence for which models might actually work – and which remain wishful thinking.
The Walled Garden Approach looked more plausible at the time of that post than it does now. I suggested publishers could license catalogues to AI platforms where readers generate personalised story variations that stay within the platform. But the music walled garden works because fans want infinite variations on beloved songs, and WMG’s insistence on download restrictions and caps specifically addresses AI tracks flooding streaming services. The threat music labels face is displacement – AI-generated content competing with human-created music on the same platforms.
Books face a different problem. AI-generated story variations aren’t likely to flood Amazon or physical bookstores, contrary to our industry’s home-grown urban mythology. We have collectively elevated the sky is falling to a fine art.
Rather, the displacement risk is that AI models trained on books can generate competing commercial content – full novels, not variations – without publisher involvement. A walled garden that keeps AI content contained doesn’t solve this problem because the real competition happens outside the garden entirely.
The Remix Licensing Model – character AI chatbots, story continuation tools, perspective shifting – seems more viable but faces the economic sustainability question I raised initially. Music is repeated consumption where people listen to favourite songs hundreds of times, whilst books are typically read once, maybe twice. Would readers pay recurring fees to interact with AI versions of beloved characters? Some would, certainly. Enough to sustain the kind of revenue sharing that makes partnership worthwhile for major publishers? That’s far less certain.
The Collaborative Franchise Model I proposed – where AI enables official “expanded universe” content with algorithmic filtering and semi-official canon status – now seems most realistic, but only for properties with established franchise value. Think Harry Potter, Hunger Games, major fantasy series. For the vast majority of books that aren’t multimedia franchises, this model offers little.
The music settlements reveal something crucial: partnership works when platforms need content they can’t replicate and when that content has massive existing fan engagement. Suno valued at $2.45 billion signals millions of users wanting to remix Taylor Swift and BTS. Would a “Harry Potter AI fiction platform” attract comparable engagement? Absolutely! Would a “general literary fiction AI remix tool” licensed by mid-list publishers? Almost certainly not.
The Emerging Two-Tier System
What’s crystallising across both music and book publishing is a two-tier system with fundamentally different dynamics.
Tier One: Properties with irreplaceable value. Major music artists whose voices and performances can’t be legally replicated. Book franchises with massive existing fan engagement and multimedia presence. These properties have genuine leverage to negotiate valuable partnerships because AI platforms need them to attract users and because their owners can credibly threaten disruption through litigation.
Tier Two: Everything else. The thousands of artists whose music helps train AI models but doesn’t individually drive platform adoption. The hundreds of thousands of books that contribute to AI training but don’t have Harry Potter-level fan engagement. These creators receive compensation primarily through class action settlements that address acquisition methods, not through ongoing collaborative partnerships.
The music settlements all involve major labels representing marquee artists. The book settlement involved authors collectively through class action. This distinction matters. Individual leverage concentrates value capture. Collective leverage spreads it thin.
The Case for – and Against – Going to Trial
My earlier analysis suggested both sides recognise uncertainty, making settlement rational. The pattern since then reinforces this but adds a crucial element: timing pressure.
Robert Kyncl framed WMG’s settlement as seizing “this opportunity to shape models that expand revenue” whilst “Suno rapidly scales, both in users and monetization”. The subtext is clear: settle now whilst negotiating power exists, or watch AI platforms become so entrenched that they dictate terms. Music labels aren’t settling because they fear losing at trial – they’re settling because delayed victory may be worthless if platforms establish market dominance first.
A prisoner’s dilemma for book publishers.
This creates a prisoner’s dilemma for book publishers. If major publishers coordinate to pursue litigation through appeals and establish strong precedent, they could potentially secure better long-term positioning. But if even one major publisher breaks ranks to pursue individual partnerships, others face pressure to follow before terms worsen further. Sony’s continued litigation against Suno and Udio represents this holdout position – but if those cases also settle, it confirms that coordination is unsustainable.
For book publishers, the case for trial rests on a few key factors:
- The fair use question needs clarity. The Copyright Office concluded in May 2025 that “it is not possible to prejudge litigation outcomes” and that “some uses of copyrighted works for generative AI training will qualify as fair use, and some will not”. Appellate precedent could narrow or expand fair use in ways that dramatically affect publishers’ leverage. Settlement foregoes this potential advantage.
- Market-wide licensing schemes require legal pressure. Copyright lawyer Cecilia Ziniti called the Anthropic settlement “the beginning of a necessary evolution toward a legitimate, market-based licensing scheme for training data”. But this only works if AI companies fear litigation consequences. If settlements become cheap insurance rather than meaningful concessions, licensing may never materialise.
- Fragmentation creates race-to-the-bottom dynamics. Without coordination or legal precedent forcing industry-wide standards, individual publishers negotiating with individual AI companies will likely accept progressively worse terms. Trial offers the possibility of precedent that applies universally, rather than bespoke agreements that vary by negotiating power.
The case against trial is equally compelling:
- The core legal question is already lost. Training on legally obtained books is fair use according to existing precedent. Taking cases through appeals might affirm this rather than reverse it, leaving publishers with both litigation costs and weakened bargaining position.
- Platform entrenchment is happening now. Suno’s $2.45 billion valuation and rapid scaling mean delay has costs. If publishers wait for final appellate rulings whilst AI platforms integrate into content creation workflows, subsequent partnerships may offer less value than deals struck today.
- Negotiated terms beat adverse judgments. Even unfavourable partnership agreements that generate modest revenue streams exceed outcomes where publishers receive nothing because courts rule comprehensively for fair use.
- The music precedent creates momentum. As more content industries pursue partnership over litigation, courts may view adversarial approaches less sympathetically. Being the last holdout risks judicial perception that publishers are unreasonably refusing reasonable licensing terms.
What Publishers Should Actually Do
Synthesising these dynamics suggests a strategic framework rather than binary choice between litigation and settlement:
Coordinate on minimum standards whilst pursuing selective litigation. Rather than industry-wide litigation or universal settlement, publishers should establish baseline principles -attribution requirements, compensation frameworks, opt-out mechanisms – that any AI partnership must include. Simultaneously, pursue litigation on specific issues where precedent could strengthen negotiating position, particularly around output infringement and non-transformative commercial uses.
Differentiate by property value. Publishers controlling major franchises should negotiate individual partnerships from strength, similar to music labels with marquee artists. Publishers with extensive backlists but no franchise properties should pursue collective licensing through organisations like the Association of American Publishers, creating economies of scale that individual negotiation can’t achieve.
Invest in platform alternatives. Rather than licensing to AI platforms on unfavourable terms, publishers should explore cooperative AI tools that keep value within publishing. Imagine a publisher-owned AI platform that enables legitimate fan engagement with licensed properties whilst returning 80-90% of revenue to rights holders rather than the 20-30% typical of technology platform deals.
Focus on output infringement, not training. The Anthropic settlement doesn’t cover future claims based on infringing LLM outputs, preserving this avenue for litigation. Publishers should develop sophisticated monitoring of AI-generated content and aggressively pursue cases where outputs constitute clear infringement, establishing precedent that training fair use doesn’t immunise derivative infringement.
Accept the two-tier reality. Stop pretending every book has leverage equivalent to Taylor Swift’s discography. Properties with genuine franchise value should extract maximum partnership terms. Everything else should pursue collective licensing or accept that training is fair use whilst focusing on output policing and discovery advantages (AI-generated content might actually expose readers to human-authored books they’d otherwise never encounter).
The Uncomfortable Truth About Leverage
Perhaps the hardest truth the music settlements reveal is how much leverage actually matters – and how little book publishing has in comparison.
WMG described Suno as “rapidly scaling, both in users and monetization,” positioning the settlement as seizing an opportunity. But this frames leverage backwards. Suno could continue scaling without WMG, training on legally purchased music, accepting litigation risk, and betting on fair use precedent. WMG couldn’t prevent this except through injunctions that courts seem increasingly reluctant to grant before final judgments. The leverage isn’t “Suno needs our music” – it’s “we can make Suno’s growth more expensive and uncertain until they pay us to go away.”
AI companies can train on legally purchased books right now, with existing precedent supporting fair use.
For book publishers, even this leverage is weaker. AI companies can train on legally purchased books right now, with existing precedent supporting fair use. They can scale without partnerships. The only leverage publishers retain is making this slightly more expensive through litigation and potentially constraining some outputs through infringement claims.
The earlier observation that x% of something beats 100% of nothing assumes something is available. For many publishers, especially those without franchise properties, the “something” on offer may be so minimal that strategic alternatives – investing in author discovery, focusing on physical book experiences that AI can’t replicate, developing proprietary publishing AI tools – generate better returns than marginal licensing revenue.
No Gunfight At The AI Corral, But No Victory Either
The music industry’s settlement cascade confirms my core insight from the earlier post: no one’s willing to risk the gunfight at the AI Corral when negotiated peace offers acceptable terms. But “acceptable terms” means something profoundly different for music than for books.
Music settled from strength into ongoing partnerships with major commercial platforms. Books settled from weakness into one-time payments for past wrongs whilst the fundamental fair use question was decided against them. Music labels are shaping AI integration. Book publishers are responding to it.
My earlier analysis concluded by asking whether publishers will help shape AI evolution or simply respond to it once others have defined the landscape. Six weeks later, the answer emerges with uncomfortable clarity: most publishers will respond, not shape. A few – those controlling genuine franchises – may negotiate meaningful partnerships. The rest will collect modest settlements, watch AI platforms train on legally purchased books, and hope that human authorship remains commercially viable as AI-generated content becomes ubiquitous.
The safe settlement option I’ve identified exists for music because both platforms and labels had something the other needed. For book publishing, platforms increasingly don’t need what most publishers have. Thanks to an ill-advised class action that far too many in the industry rallied behind without thinking through the consequences – they can get it legally for retail price. That’s not a negotiating position. It’s a fait accompli dressed up as collaboration.
The real question isn’t whether settlements beat courtroom confrontation. It’s whether publishers have anything left to settle for
This post first appeared in the TNPS LinkedIn newsletter.
