The content published on the internet has become part of the infrastructure that artificial intelligence companies rely on to develop their models and provide answers to users' questions. However, this use raises controversy over the boundaries between making information available to the public and holding the right to copy or exploit it commercially.
The debate is not only about the ability of tech companies to access web pages, but also about the rights of content creators, the mechanisms for compensating them, and the impact of new services on the economy of digital journalism.
The process of developing artificial intelligence models relies on large datasets of textual and other data, which are processed to extract patterns and relationships that help the models produce answers. Automated crawling programs are used to collect online content, navigating between pages and links to extract available texts and data, before the materials undergo cleaning, organization, and processing.
OpenAI explains in its official statement on its data handling approach that the sources for model development include publicly available information on the internet, data obtained through partnerships with other entities, and information provided by users or trainers under relevant conditions. The diversity of sources means that data collection does not necessarily occur in the same way among all companies or at all stages of model development.
Data collection processes raise legal disputes between tech companies and publishers over copyright and terms of use (Getty), but technical access to a published page alone does not resolve the issue of the legal right to use it. For example, the ability to read a news article without a subscription does not necessarily mean that its author has waived copyright or granted permission for its reuse for any commercial purpose.
The answer varies based on the applicable law, the nature of the material, the way it is used, and the surrounding circumstances. In the United States, some uses are subject to the fair use doctrine, which requires a legal assessment that takes into consideration multiple factors, including the purpose and nature of the use, the nature of the protected work, the amount used, and the effect of the use on the potential market for the original work.
The U.S. Copyright Office has dedicated part of its formal study on artificial intelligence to the issue of training models on copyrighted materials, indicating that the issue involves legal and policy questions that cannot be reduced to the content being publicly available. This does not mean that all training uses are illegal, or that all of these uses are automatically covered by fair use; the ruling depends on the specifics of each case and the law applicable to it.
The issue becomes more complex when it comes to news articles, images, and paid archives, or when AI systems use protected material to provide summaries or answers that might make some users forgo visiting the original source. Here, the question of the legality of data collection intersects with questions of economic competition and the impact of use on the value of journalistic work.
The disagreement takes two overlapping paths: the first relates to the right to use protected materials in training models, and the second deals with how the content generated by these models is presented and its impact on the commercial interests of publishers.
In a development reflecting the ongoing dispute, Reuters reported a few days ago that media organizations, including The New York Times, Agence France-Presse, and The Economist, have joined a coalition aimed at establishing technical standards to monitor the use of content by AI companies and facilitate licensing.
The agency also referred to lawsuits filed by media entities against tech companies, including a suit accusing OpenAI of copyright infringement. These are accusations and legal claims, not definitive judicial rulings in themselves.
Conversely, some companies and publishers are turning to commercial agreements instead of resolving all disputes through the courts. Reuters also reported that CNN, Financial Times, Axios, and other publishers have joined Google's AI News program, which allows their content to be used in generative features in exchange for payments, while providing attribution and direct links to sources.
The financial terms of the agreements have not been disclosed. These agreements indicate that licensing may represent a commercial pathway for regulating use, but it alone does not prove that all unlicensed use is illegal, nor does it mean that every agreement covers all types of use or all published materials.