Meta drops contract clause that let it scrape news sites for AI training
After backlash, Meta has removed language from its commercial terms that appeared to permit using content from any web domain or URL to develop and improve its AI.
Meta has removed language from its commercial terms page that appeared to give it broad permission to pull content from news sites and other web domains for AI development. The clause stated that Meta could use content from such domains, web pages or URLs to provide, develop and improve its products and services. It no longer appears on the page after the change drew criticism.
The wording mattered because of how Meta's terms are structured. Commercial terms apply to businesses and publishers that interact with Meta's platforms, and a clause covering "any content" from URLs reads far wider than a typical licensing agreement. Publishers and media organizations have been watching terms of service closely since AI training data became a contested commercial asset, and language that skirts explicit licensing is exactly what they flag.
This is the second time in recent months that Meta has rewritten terms after an outcry. The company has been under scrutiny for how it sources material for its AI models, including Llama, while simultaneously negotiating content deals with some publishers. Removing the clause does not by itself settle the question of what Meta actually ingests; it only narrows what its public contract claims it may do.
The timing is notable. AI developers face a growing set of lawsuits and regulatory probes over training data, and the European Union's AI Act adds disclosure obligations for general-purpose models. A broad scraping clause in a public contract can be used as evidence of intent in those disputes, which gives Meta a reason to tighten the language even if its internal practices don't change.
What to watch is whether Meta publishes a replacement clause with narrower scope, and whether other platforms follow with similar edits. For publishers, the practical question remains unchanged: the terms no longer say Meta can scrape, but nothing in the revision confirms it won't. Rights holders will likely press for explicit opt-outs, and any future licensing framework will be judged against what these terms said in the interim.