Meta revised its business terms for accounts, removing explicit permission to train AI on content from associated news sites. The changes were announced ahead of an October 30 effective date. The original language had said Meta may use any content from a business account’s website to develop, research, and train its products.
Legal scholars argue the revised wording still leaves room for Meta to scrape entire sites. Cornell Law School professor James Grimmelmann told Nieman Lab that the clause “associated with your products or services” can be interpreted to permit broad data collection, potentially overriding standard web-crawling restrictions such as robots.txt files.
Media-industry attorneys echo the concern. Matt Topic of Loevy & Loevy explained to Nieman Lab that the term “develop any business product” is undefined, and because Meta is building AI-related offerings, “develop” could be read to include training AI models significantly on the harvested material.
The initial draft of the terms had sparked a strong reaction from publishers, who feared that Meta could claim its commercial agreements trumped any site-owner instructions to block scraping. Critics warned that such a stance would undermine the protective function of robots.txt and similar mechanisms that limit automated access.
After the backlash, Meta removed the explicit reference to “training” but retained language that allows use of scraped content for product development. Observers note that the lack of a precise definition for “business product” may still enable the company to justify AI training activities under the revised agreement.
The debate highlights broader tensions over how large platforms acquire data for generative AI. Publishers argue that unconsented scraping threatens copyright and revenue, while platform lawyers contend that existing terms grant sufficient rights. The outcome may influence future contractual language and regulatory scrutiny of AI data practices.