Newspaper publishers rage against Meta's new deal - concerns about being exploited for AI

On October 30, Meta's new user agreement for business customers comes into force. The agreement is mandatory and applies to all companies that operate, for example, Facebook pages or Instagram profiles.

When it was first presented, many people went crazy. One clause appeared to give Meta the right to pluck information from the affiliated companies' sites and use it to train its AI systems. It stated that Meta may use "all material" from the websites to "develop, improve, research, train" its products. Legal experts interpreted this as Meta taking the right to, for example, all articles on a news site for its own purposes.

- For news publications, this clearly means permission for Meta to train its AI on any content from their sites, perhaps even using it to compete with the publishers themselves, James Grimmelmann, a professor of information law at Cornell Law School told the Nieman Lab media site.

Now Swedish media organizations are also protesting against Meta.

- Journalism is not a free product that can be used however you like. If Meta wants to use the media's journalistic content, it must be done through voluntarily entered into agreements and on market terms. Meta's actions are an abuse of the company's market position, says James Savage, chairman of Utgivarna, which represents Swedish media, in a press release.

The industry organization Tidningsudgivarna demands that the Swedish authorities review the agreement.

- Meta needs to clarify how the new conditions are to be interpreted and used, says Tidningsudgivarna's Johan Taubert in a press release.

The criticism has also been sharp internationally. The World Association of News Publishers, which represents over 20,000 media brands in 120 countries, has in a letter called on Meta to stop the new terms of use.

The criticism has already prompted Meta to make some changes. The wording about training AI systems with data from companies' own websites has been replaced with new wording. It says nothing about specifically training AI systems, but rather that Meta can retrieve information to "improve" and "maintain" advertising and business products.

A Meta spokesperson told Nieman Lab that the companies can stop the retrieval of articles and other information through robots.txt, an industry standard that allows site owners to tell search engines and AI companies which parts of a website are allowed to use. However, it does not stop the AI ​​companies technically, but is based on them voluntarily obeying the instruction. News sites should be able to stop retrieval through robots.txt, Meta promises. "That's always been the case and it hasn't changed," a Meta spokesperson told Nieman Lab.

But that has not silenced the critics. Professor James Grimmelmann remains skeptical and believes that Meta has chosen such broad formulations that they do not limit Meta's possibilities very much. Others note that the terms do not mention companies' ability to limit Meta through robots.txt.

The issue of rights around training data for AI systems has been a contentious topic for years. As the AI ​​boom took off in the fall of 2022, when Chat GPT was released, many of the AI ​​companies started training their models on anything they came across, usually without asking permission.

- Today we train them on what is publicly available, Google CEO Sundar Pichai admitted in a DN interview in 2023.