Meta's updated commercial terms of use carried a provision that the company can "use any content from [a business account's] domains, web pages, or URLs to provide, develop, improve, research, train, and maintain the integrity of" Meta's products and services.
Nieman Lab took that wording to legal experts, who called it a "remarkable provision". Read plainly, any publisher that held a Meta business account had granted Meta permission to scrape its site and use the text to train its models — a much wider grant than the ad-targeting language the terms were ostensibly about.
Meta's response was that nothing had actually changed. "We've heard you and want to clarify: the crawling terms we consolidated only applied to our ads and business products, and we've clarified that in the terms themselves," a spokesperson said. "It does not change how Meta uses web crawlers."
Then The Verge reported that the provision had been removed from the terms altogether.
Why it matters
- Terms of service, not blog posts, are where AI training permissions actually live. A single clause buried in a commercial agreement is the difference between a crawler being welcome and a crawler being a trespasser.
- The wording was consolidated from older documents, which is how this kind of clause usually arrives: nobody writes "we may train on your site" on purpose, it accretes during a cleanup.
- The removal is a statement about intent, not a change in behaviour. Meta says its crawling never depended on the clause, which means the protection publishers thought they had in the old text was thinner than it looked.
- For anyone running a site, the lesson is to read consolidated terms as new terms. Consolidation is where scope quietly expands, and it only gets walked back when someone notices.