The journalists’ revolt against AI scraping

In just a few short years, artificial intelligence (AI) tools such as ChatGPT and Claude have become mainstream technologies. The companies behind these innovative products tout their immense potential benefit for humanity, but this narrative, while not necessarily wrong, brushes aside that these models are built on the wholesale theft of copyrighted work and threaten to weaken further an already-fragile pillar of democracy: journalism.

That is why Editorial Perfil, a major Latin American publisher headquartered in Argentina (and the owner of Perfil, the newspaper where I am digital director), recently sued ChatGPT’s owner, OpenAI, and the company’s largest external shareholder, Microsoft, for the improper use of Editorial Perfil’s journalistic content in their AI products. In the damages suit brought before the Argentine Federal Civil and Commercial Court, we also accused these companies of “unfair competition” — in other words, abusing their oligopolistic position to our detriment.

While Editorial Perfil is the first Spanish-language publisher to file a copyright-infringement suit against an AI firm, it follows in the footsteps of the New York Times, which sued OpenAI and Microsoft over the same issue in late 2023. The Brazilian newspaper Folha de Sao Paulo brought a similar case last year, resulting in a content-licensing agreement earlier this year (an approach that OpenAI has taken with other publishers, including PRISA, the Spanish conglomerate behind El Pais).

We have been here before: Dominant search and social-media platforms also benefited from copyrighted news content without always sharing in the spoils. Search engines — such as Google’s “free” model — made news articles immensely more accessible by featuring them on their results pages. However, despite relying heavily on this copyrighted content to boost ad revenue, Google was not required to provide adequate compensation to the media companies that might employ hundreds of journalists to produce it. True, these distribution systems drove traffic to publishers’ Web sites, but the playing field was never level and increasingly tilted toward tech firms as they consolidated their power.

In the AI age, OpenAI and Anthropic (which owns Claude), as well as the entrenched tech giants such as Google and Meta (which owns Instagram and Facebook), have supercharged this dynamic. The chatbots and other AI enterprise products they have built on the back of increasingly sophisticated large language models answer Internet users’ questions directly, depriving news outlets of traffic and the chance to serve ads or sell subscriptions to potential readers.

These models required huge amounts of training data, much of which was copyrighted material illegally scraped from the Internet. Anthropic recently settled a class-action lawsuit in California by agreeing to pay US$1.5 billion to book authors who argued that the company had downloaded pirated copies of their works to train Claude. The models constantly thirst for more, particularly up-to-date information on current events.

At Editorial Perfil, we have identified multiple instances in which ChatGPT can replicate content from our historical archive in detail. It does the same with real-time news and would even prompt users to request a daily news recap so that they can avoid wasting time visiting multiple sites. It easily bypasses paywalls; for articles that it cannot find within Editorial Perfil’s own sites, it would use copycat Web sites that illegally reprinted them. Historical server data show that multiple bots are consistently crawling our content — not just from OpenAI, but from all the AI firms as well.

These flagrant abuses are self-evident to anyone with common sense. AI firms are not only stealing content, but also encouraging users to stay within their closed ecosystems, turning the Internet into an oligopolistic market. Yet they still try to argue that they are engaging in “fair use” of “publicly available” content. Unfortunately for OpenAI, it has already acknowledged the value of journalism through its many licensing deals. By contrast, Anthropic has avoided content-licensing agreements with news organizations.

The news industry has been in crisis ever since the digital revolution upended its business model more than two decades ago. A new, more sustainable model never emerged. While those of us in media failed to anticipate how the rise of the Internet would change everything, we tried to reinvent ourselves after the fact. However, the oligopolistic dominance of “big tech” has prevented us from making any real progress, and AI is exacerbating the crisis.

A decimated news industry is a bad outcome for everyone. It would not only result in a less informed public; it would also harm the parasitical AI firms themselves, cutting off an important source of training data. If they cannot be bothered to act in the public’s interest, they should at least look out for their own.

Agustino Fontevecchia is digital director at Perfil in Argentina.