On July 20, 2026, a U.S. federal court finally approved the settlement in which Anthropic will pay $1.5 billion to resolve a class-action lawsuit brought by authors and copyright holders. This is one of the most relevant economic agreements recorded so far in copyright matters.
The news has been summarized as a defeat of artificial intelligence against the authors. However, the case is much more complex and its true importance lies precisely in that complexity.
The court had considered that the use of legally acquired books to train models could constitute fair use under US law.
What it did not accept was that Anthropic downloaded and stored millions of copies from pirate libraries to build its repository. The agreement, therefore, does not mean that all training with protected works is automatically illegal. It means something equally important: the origin, traceability and legitimacy of data matters.
For years, part of the technology industry acted under a simple logic: first collect all the available information and then discuss whether there were rights over it. The Anthropic case shows that that stage is ending.
When a company uses these creations to develop a commercial product, intellectual property ceases to be an exclusively cultural or legal discussion. It becomes part of the cost structure, risk management and the business model itself.
A warning for Latin American startups
A startup that trains a system with information whose origin it does not know does not only have a legal problem. You have an investment problem. What will happen during a funding round when the investor asks where the data comes from? Can the company prove that it had the right to use them? Do contracts with developers, researchers and suppliers adequately transfer rights? Are there restrictions on marketing the model in other markets?
Not having those answers can affect the company’s valuation, block an acquisition or transform a promising product into a difficult-to-sell liability. In the European Union, providers of general-purpose AI models must implement copyright compliance policies and publish a sufficiently detailed summary of the content used to train their models. These obligations can even reach suppliers established outside Europe who market their technologies there.
This means that data traceability will progressively cease to be a voluntary good practice. It will be a condition to access markets, capital and institutional clients.
Copyright is not the enemy of innovation
Protecting is the possibility of obtaining an economic return is precisely what encourages people and companies to continue researching, creating and innovating. Without protection, innovation turns into extraction. With protection, it becomes a sustainable knowledge economy.
The real challenge of artificial intelligence is not to eliminate copyrights, but to build models capable of respecting them, managing them and generating value for all those who participate in the innovation chain.
