Press and AI: Licenses and Fair Use in Court

Key Takeaways

1Several copyright infringement lawsuits against OpenAI and Microsoft are consolidated, and the discovery phase is nearing its end

2Publishers, including The New York Times and Ziff Davis, report a decline in traffic and the dissemination of AI-generated content

3OpenAI invokes fair use and implied licenses via robots.txt, citing its partnerships with the press

4The debate over costs contrasts Vivek Shah, who considers the amounts manageable for AI, with the investments announced by OpenAI

Why it matters : The outcome of these lawsuits could define the rules for access and compensation for content used to train generative AIs.

As the discovery phase comes to a close and complaints against OpenAI and Microsoft have been consolidated, publishers are demanding compensation for the training of AI models. OpenAI invokes fair use and implied licenses via robots.txt. A numbers game pits the press against Vivek Shah and OpenAI, while blocks on robots and paywalls are multiplying.

Stricter blocks and paywalls announced by publishers

Many publishers are now blocking access for OpenAI's robots and those of other AI companies to their sites. Vivek Shah believes that these barriers and paywalls will strengthen, which would limit access to information and hinder an open web. He attributes this development to the uncontested idea that licensing would be difficult and costly.

Procedure: discovery phase nearing its end and key questions ahead

The discovery phase is nearing its conclusion in the copyright cases against OpenAI, with several distinct complaints, including some against Microsoft, consolidated before a single judge. The parties submitted requests this month for the federal court to rule on points such as fair use. The judge may decide on some major questions before sending the cases back to their original districts for potential trials. Thus, all these lawsuits are entering a new phase.

What publishers are complaining about: declining traffic and AI content

Among the plaintiffs are The New York Times and Ziff Davis, owner of CNET. In a joint document, several newsrooms assert that the use of their articles to train ChatGPT has reduced their web traffic and flooded the market with AI-generated content described as "pink slime." They also link the release of ChatGPT in November 2022 to the acceleration by other players, including Google, of generative AI products such as AI Previews.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

OpenAI's defense: fair use and implied license

OpenAI argues that training on data collected online falls under fair use, as the use would be highly transformative, focusing on very factual works, with no recognizable harm to the market. The company also claims an implied license when publishers did not block access via robots.txt or similar means. It points out that search engines have long copied web pages to create indexes, that the plaintiffs use their own scrapers, and that this copying is understood as implicitly authorized unless opted out. OpenAI further highlights partnerships with news organizations and presents its systems as beneficial for journalism and the public.

The cost debate: $20 billion vs. $750 billion in investments

The federal government has indicated that licensing requirements could weigh on AI companies, an intervention framed in the context of the Trump administration. Vivek Shah contests this view by citing orders of magnitude: less than $20 billion per year in royalties in the music industry in the U.S. versus $750 billion that OpenAI has indicated it wants to allocate to infrastructure by 2030. According to him, around $20 billion per year in licenses for publishers would be negligible for AI players while being crucial for the press, and consistent pricing would promote the quality of both incoming and outgoing content.