Meta Reverses Privacy Terms Update After Scrutiny Over Scraping and AI Training
Meta removed language from its privacy policy that critics said would have made it harder to object to web scraping and AI training, insisting its crawling practices never changed.
Meta has pulled language from its privacy terms that critics argued could have expanded its ability to scrape the open web and train AI models on the resulting data. The company confirmed the reversal after the wording drew scrutiny, with a spokesperson saying the edit never reflected a shift in how Meta actually crawls the web. The change came and went quietly, but the underlying question it raised is not going away.
The disputed language sat inside Meta's privacy policy, the document that governs how the company collects and uses data across Facebook, Instagram and its other products. According to TechCrunch, the update was read by outside observers as loosening the basis on which people could object to automated collection of public web content, a category that increasingly feeds large language models. Meta's position is that the wording was misunderstood: the spokesperson maintained that the revised terms did not alter its crawling behavior.
That distinction matters because the gap between what a policy says and what a company does is exactly where AI training disputes get litigated. Public web scraping has become one of the most contested practices in the industry, with publishers and site owners arguing that content pulled for model training amounts to uncompensated use of their work. For Meta, which is building out its Llama model family and an array of AI features across its apps, access to large volumes of text and images is a direct competitive input.
Privacy and content advocates have grown used to reading policy diffs closely, and this episode shows why. Terms of service are often the only public signal of how a platform intends to treat data, and a few altered sentences can shift what users and website operators believe they have consented to. When a company walks back that wording after pushback, it leaves open the question of whether the original draft was sloppy drafting or an intentional expansion that proved too costly to defend.
The reversal also lands against a backdrop of regulatory pressure on both sides of the Atlantic, where data protection authorities and AI legislation are scrutinizing how training corpora are assembled. Meta is not alone in facing this scrutiny; several major AI developers have been challenged over whether their data collection practices align with the terms they publish. A spokesperson's assurance that nothing changed operationally will not settle the broader argument over what should be permitted.
What to watch next is whether Meta publishes clearer, more durable language about automated collection and model training, or simply lets the issue fade. Site owners and publishers who rely on robots.txt and terms of service to control their content will be checking whether the final policy matches the practice. The answer will shape how much of the public web is treated as fair game for training data.