Meta has removed a contentious provision from its commercial terms after legal experts raised concerns that the wording could be interpreted as giving the company broader rights to use third-party website content, including news articles, to train its artificial intelligence (AI) models.
The provision was removed on October 17 following scrutiny over language that appeared to permit Meta to use content from specified domains, webpages and URLs to develop, research and train its products and services.
Critics argued that the wording could potentially extend beyond content that website owners had authorised for such purposes, raising questions about copyright, consent and control over online material.
Meta has denied that the change reflects a shift in its web-crawling practices, maintaining that the revised terms do not override website owners’ preferences.
Why Meta’s commercial terms came under scrutiny
The controversy centred on language in Meta’s commercial terms that referred to the use of website content for developing and training its products. Legal commentators, including Cornell University professor James Grimmelmann and experts associated with the Nieman Journalism Lab, raised concerns about how broadly the provision could be interpreted.
According to reporting by The Verge, the language has since disappeared from the relevant page. Critics suggested the clause could be read as permitting Meta to collect website content for AI training without compensating publishers, while potentially taking precedence over restrictions used by website operators to control automated access.
One mechanism at the centre of the debate is robots.txt, a file through which website owners communicate instructions to automated crawlers about which parts of their sites can be accessed. However, the legal effect and enforceability of these instructions can vary depending on the circumstances.
The concerns related to the possible implications of the wording, rather than establishing that Meta had scraped news websites in violation of publishers’ preferences.
Meta rejected the suggestion that the revision granted it new rights over third-party content. A company spokesperson said the consolidated crawling terms applied to its advertising and business products and that the company had clarified their scope.
The spokesperson also maintained that website owners could continue using robots.txt to communicate their preferences and that the revised wording did not supersede those instructions.
AI training data remains a contentious issue
Meta’s decision comes amid growing tensions between technology companies seeking large volumes of data to train AI models and publishers, creators and digital rights advocates concerned about how their work is collected and used.
Meta already uses certain publicly shared content posted by adult users on its platforms to improve its AI systems. Elon Musk’s xAI also uses public posts on X to train its Grok chatbot. However, the public availability of online material does not, by itself, settle questions about copyright, consent or the conditions under which that content can be used for AI development.
Meta has also faced legal challenges over its use of books to train AI models. In 2025, a US federal court ruled in the company’s favour in a copyright lawsuit, finding that the plaintiffs had not established that Meta’s use of their works was unlawful. The ruling did not resolve all disputes over AI training, and litigation involving Meta and other AI companies continues.
The episode echoes a similar controversy involving file-transfer service WeTransfer. In July 2025, the company faced backlash after an update to its terms appeared to permit the use of uploaded files to improve machine-learning models. WeTransfer subsequently revised the wording and removed references to AI and machine learning.
Meta’s latest revision clarifies the scope of its commercial terms but leaves the wider debate over the collection and licensing of online content for AI development unresolved. For publishers and website operators, the distinction between material that is publicly accessible and content authorised for AI training remains a central concern.