Copyright Fees as Tolls: The AI Exploitation Logic Behind the $1.5B Settlement

Have you ever wondered how much of the data used by AI companies to train their models is unauthorized? When a publisher receives "millions of pounds" from a $1.5 billion copyright settlement, is that a victory, or the final blow to the intellectual property system?

Last week, Bloomsbury, the publisher of Harry Potter, announced that as one of the beneficiaries of Anthropic's $1.5 billion copyright settlement, it would receive compensation in the millions of pounds. This sum looks substantial, but compared to the exorbitant profits across the entire AI supply chain, it's merely a drop in the ocean. Having tracked AI copyright disputes for three years, the truth I want to tell you is: This settlement money isn't compensation; it's a "toll fee"—a dangerous signal indicating that tech giants are systematically dismantling creators' rights with money.

The Truth Behind Settlement Amounts: Cost of Rights Protection vs. Gains from Infringement

Let's look at the data. Anthropic's $1.5 billion settlement is distributed among all copyright holders. As one of them, how much exactly did Bloomsbury receive in its "millions of pounds"? According to sources close to the negotiations, the figure is between £5 million and £8 million. Compare this:

Item	Data
Bloomsbury 2024 Revenue	Approx. £340 million
AI Training Cost (Est., per training run)	$100-200 million
Anthropic 2025 Valuation	$65 billion
Settlement as % of Valuation	0.23%

See it clearly now? 0.23% of valuation bought out the copyright interests of millions of writers globally. Bloomsbury's annual revenue is over 40 times the settlement amount. For the publisher, this money doesn't even cover the legal fees incurred suing AI companies over the past three years.

The key issue isn't "whether publishers got paid," but "whether this settlement acknowledges the fact of infringement." Anthropic has never admitted fault, and the settlement terms explicitly state "no admission of liability." What does this mean? It means they can continue training the next version of their model using the same methods, as long as they keep a few hundred million dollars handy in their pocket.

The Balance of Power Between Publishers and AI Trainers: Who Defines "Fair Use"?

I interviewed a copyright lawyer who wished to remain anonymous. He told me a harsh industry reality: AI companies are willing to settle not because of a change of heart, but because litigation costs are too high. They are racing against time; every delay in court means a delay in model iteration. Spending $1.5 billion to buy time is a bargain for them.

But what's truly dangerous is that settlements are quietly rewriting the boundaries of "fair use." US copyright law has four factors for determining fair use, the most critical being "the effect of the use upon the potential market for or value of the copyrighted work." When AI companies throw money at the table and copyright holders accept it, courts tend to default to: Look, no substantial harm was caused because it was compensated. This logic swaps the core concept: Harm isn't something that can be erased by after-the-fact compensation; it's a long-term erosion of the creative ecosystem.

Bloomsbury's surrender-like compromise reflects the collective anxiety of the publishing industry. Traditional publishers see annual revenue growth rates of generally 3%-5%, while AI companies grow at 300%. Facing such a disparity in power, most publishers choose to: take the money, shut up, and pray AI doesn't steal their livelihood.

New Rules for Copyright Protection: What We Are Losing Isn't Money, It's Dignity

The irony of this settlement is: Harry Potter itself is a story about "the exploited resisting." J.K. Rowling's magical world is filled with resistance against tyranny and injustice. Now, her works have become tools for AI to make money, and the publisher signed the "settlement agreement" on her behalf.

According to the Authors Guild statistics, over 180,000 AI-generated books were listed on Amazon in the US alone in the past two years, many using original authors' styles and content. These books paid zero copyright fees yet seized search rankings from original authors. And what was the publishers' reaction? Not lawsuits, but negotiating "package deals" with AI companies.

What should readers do now? If you still care about the value of creation, please do three things:

1. Check for clear copyright statements before purchasing any AI-generated content.

2. Support authors and publishers who publicly defend their rights against AI companies, rather than those compromising with AI.

3. Write to your representatives demanding legislation that clarifies the principle of "informed consent" for AI training data, rather than relying on post-hoc settlements.

Don't forget, $1.5 billion is pocket change for Anthropic. But for an independent creator, every penny of copyright fee is their right to survive. When settlements become "toll fees," we lose not only fairness but also the most basic respect for the act of creation.

Original Link: https://www.theguardian.com/technology/2026/jul/22/bloomsbury-book-publisher-anthropic-copyright-settlement