Meta Drops Wording That Worried News Sites About AI Copying

Meta has taken out wording that seemed to let it copy outside websites, including news sites, to train its AI. The Verge reported the change on Oct. 17.

The removed sentence said Meta could use any content from domains, web pages, or URLs to run, develop, improve, research, train, and maintain its products. That sentence is now gone from its commercial terms page.

Spokesperson Daniel Roberts said that wording was only meant for ads and business products. He said the update does not change how Meta's crawlers work, and does not give Meta new rights that overrule a site owner's robots.txt preferences. In practice, what a crawler does depends on its code, robots.txt signals, and any separate contract permission. Robots.txt is the small file a website uses to tell automatic collectors what to leave alone.

Roberts had earlier disagreed with reporting on the update. The Verge described that disagreement on Oct. 16, after Nieman Lab explained the draft wording.

The earlier draft, due to take effect Oct. 30, said continuing to use Facebook, Instagram, or WhatsApp would count as permission by a business account for Meta to automatically collect its website. Legal experts said that wording would allow collection from news sites, according to Nieman Lab and Editor and Publisher. That link worried news outlets that both publish a website and run business accounts on Meta apps.

Meta says the Commercial Terms changes take effect Oct. 30, 2026. Meta says they apply when you use its products for business reasons, including using ads, selling products, building apps, running a Page, running a Group for business, or using measurement tools. The Licenses part covers content shared, posted, or uploaded on or with Meta products. For U.S. people and businesses, most disputes must be settled by individual binding arbitration, with no class actions and no jury trials.

Think of it like two rulebooks getting mixed together. One rulebook covers posts or photos a business uploads directly to Meta. The other covers automatic collection from the wider web, which must follow technical signals and copyright law. Mixing them left engineering and legal teams responsible for data rules, AI records, and publisher relations without a clear line to check.

The broader context here explains the strong reaction. Site terms have become another place where AI training data is argued over, alongside robots.txt, opt-out signals, and paid licensing deals. Broad wording can cause concern even when a company says nothing will work differently in practice.

In my view, this fix reduces confusion without settling the bigger issue. Publishers still need simple, lasting ways to say yes or no to AI training. AI builders still need clear permission records that can be checked and withdrawn.

As a practical matter, read site terms like phone updates, with dates and records kept together. Check robots.txt, crawler logs, and contracts at the same time. My kids grew up assuming anything online would get copied, and for simple sharing they were mostly right. Training data is where we now sort out that assumption step by step, which should make reuse fairer for readers and builders.