Meta’s updated terms still allow AI training on scraped news sites, legal experts argue

Meta recently updated its business terms for business accounts, removing explicit language that suggested it could freely train AI models on content from news sites like The Verge. The original terms, effective October 30th, had stated that Meta “may use any content” from websites associated with a business account to “develop, research” and “train” its products. Legal experts, including James Grimmelmann of Cornell Law School, warned that this could have allowed Meta to override blocking instructions such as robots.txt by arguing that its commercial terms supersede them.

After industry backlash from publishers, Meta removed the word “train” from the terms. However, legal experts such as Matt Topic, a media and intellectual property attorney at Loevy & Loevy, argue that the revised language is still broad enough to permit AI training. Topic noted that the term “develop” is not clearly defined and could encompass AI model training, particularly as Meta develops “business products” that are widely understood to include AI systems.

Meta spokesperson Daniel Roberts stated that the update is not intended to enable scraping of news sites and that content owners can use robots.txt to signal their preferences. He emphasized that the changes do not grant Meta new rights that supersede those preferences.

The ongoing debate underscores broader concerns within the publishing industry about how AI developers access and use copyrighted content for training models. The revised terms, while narrower on their face, leave open questions about Meta’s actual practices and the enforceability of publisher restrictions.