AI and creativity: adding value and making it count

How much is human creativity worth - $1.5 billion?

That is the amount proposed by AI developer Anthropic as part of a settlement agreement to settle a multi-million dollar lawsuit brought against it by thousands of artists, creators and intellectual property rights holders. The judge has upheld the settlement. The price has been set.

Since the emergence of AI into our lives, and as a tool for mass social use, the Pandora's box of copyright has been uncovered. Indeed, generative AI has become the new sticking point between technological innovation and copyright. The cycle repeats itself, from Gutenberg to the internet, technology is shaking up our legal system. In particular, copyright.

We have already accepted and embraced it, ChatGPT, Claude or Midjourney are useful tools capable of producing high quality texts, images or music, comparable in many cases to those created by humans.

However, their functioning and quality depend on an element that is as fundamental as it is controversial: the massive use of content and works, many of them protected by intellectual property, as training "data". This is a fact. And this has been acknowledged by the heads of several AI development companies. Thousands or millions of copyrighted works have been used for training AI tools. Others call it the greatest theft of intellectual property in history. This fact has not gone unnoticed by the legislator.

As of 2 August 2025, the EU' s Artificial Intelligence Regulation (AIR) imposes a fundamental obligation on providers of general-purpose models: to respect copyright and, in particular, the opt-outs exercised by rightholders who do not wish their works to be used in training. This obligation, set out in Article 53, requires companies to implement clear technical mechanisms, machine-readable protocols, metadata or centralised registers, which make it possible to exclude reserved content. This is a substantial change, which aims to ensure that creators have effective control over the use of their works in the age of AI. Economic potential without control is useless.

The relevance of this European framework is best appreciated in light of the pact reached between Anthropic and rightholders; compensation to affected rightholders and the company's obligation to remove illegal files from its systems, although it would retain the possibility to continue to use works it had acquired legitimately.

A relevant precedent in this field is Authors Guild v. Google, in which the US courts considered that Google's mass digitisation of books, aimed at creating a search engine that displayed only limited fragments of protected works, constituted a transformative use protected by the fair use doctrine. Although Google copied the works in their entirety, the court found that the service did not replace the reading of the books or substantially harm their market, but could even increase their visibility and sales. This ruling, upheld on appeal and declared final by the Supreme Court, which refused to review the case, reinforces the idea that fair use in the United States allows for innovative uses of copyrighted works as long as they provide a different value to the original, a very different approach to that in Europe.

Here, in the European Union, the approach is much more restrictive. The system does not recognise a general principle of "fair use", but a closed catalogue of exceptions and limitations to copyright. Directive 2019/790 introduced the data mining exception, which allows works to be analysed as long as rightholders have not objected by means of an opt-out. But outside this framework, training with protected works without authorisation remains unlawful. Spanish case law has sometimes shown some flexibility, as in the Google case of 2012, in which the Supreme Court understood that displaying fragments of news in search engines was a transformative use and did not compete with the normal exploitation of the works. However, these are isolated exceptions in a system that remains much more rigid than the US system.

Beyond the legal debate, there is a fundamental issue at stake: AI models need quality content to remain useful and relevant. And that quality, for now, can only come from human creation, the fruit of imagination and the ability to project into the future. If machines were to feed only on synthetic data produced by other AIs, there would be a risk of creating a closed ecosystem, incapable of innovation. True originality remains the heritage of authors, and it is precisely this richness that the regulation seeks to protect.

In this context, it is not unreasonable to think that in the future it will be the AI development companies themselves who will act as patrons of human creators , hiring and remunerating authors, artists or journalists to nurture their models with quality content of artistic value. The tension between the two worlds is real, but so is their interdependence. The challenge will be to build a framework of respect, transparency and compensation that allows technology and creativity to coexist and enhance each other, rather than neutralise each other.

Access the full article published in El Confidencial here.