Meta's New Terms May Allow It to Train AI on News Sites for Free
The new policy allows Meta to crawl all websites 'associated' with a business account.
In a move that could have far-reaching implications for publishers and content creators, Meta has quietly updated its terms of service to permit the company to crawl and potentially use data from any website "associated" with a business account. The change, which took effect in early 2026, effectively grants Meta broad latitude to train its AI models on news content without direct compensation to the sites.
What the New Terms Say
Under the revised policy, Meta can collect and process information from any website connected to a Facebook Business account—including news sites that use Meta's tools for analytics, advertising, or content distribution. This includes pages that merely embed Meta Pixel or rely on Facebook Login for user authentication.
Industry analysts note that the wording is deliberately vague. "The term 'associated' is doing a lot of heavy lifting here," said one digital rights advocate. "It could mean anything from a business that advertises on Facebook to a publisher that simply uses Meta's comment plugin."
Implications for Publishers
The change comes as AI companies face increasing scrutiny over the use of copyrighted material for training large language models. In 2025, several high-profile lawsuits forced tech giants to rethink their data acquisition strategies. Meta's updated terms appear to be a preemptive move to secure a vast, legally defensible corpus of news content.
Publishers who rely on Meta for traffic may find themselves in a difficult position. Opting out could mean losing access to Facebook's referral traffic, while accepting the terms means their content could be used to train AI models that compete with their own journalism.
Broader Industry Context
Meta's policy shift is part of a wider trend. In 2026, AI training data has become a battleground. Google, OpenAI, and others have struck multimillion-dollar licensing deals with major news organizations. Meta, however, appears to be taking a different approach—leveraging its existing business relationships to sidestep direct payments.
Legal experts suggest that the new terms may still face challenges under copyright law, particularly in jurisdictions with strong fair use doctrines. "Just because it's in the terms of service doesn't mean it's enforceable," noted one intellectual property attorney. "Publishers will likely push back."
What's Next
Meta has not publicly announced how it plans to use the data, but the timing aligns with its ongoing development of multimodal AI systems. The company's LLaMA series of models has already been trained on web data, and the new terms could significantly expand that dataset.
For now, publishers are advised to review their Meta integrations and consider whether the benefits of association outweigh the risks of uncompensated AI training.
via The Verge AI