Meta drops AI training clause from new business terms

By VK

Published 18 Oct 2026, 09:46 UTC

Reporting from The Verge, Ars Technica, TechCrunch

After publisher backlash, Meta cut a line from its new commercial terms that experts said let it train AI on linked websites.

What it means for founders

Read terms for what a platform can take, not just what it charges. If your startup advertises or sells through Meta, the scope of crawling rights matters as much as ad prices. Read the final text before October 30, especially if your site holds paid or gated content.
Back up robots.txt with contracts. A blocking file is a signal, not a lock. If your content is your product, negotiate licensing terms directly and keep records of what you allow.
Platform risk runs both ways. Meta is pushing hard into business AI, from its new enterprise AI unit to its Muse agent, which has drawn its own privacy scrutiny. Expect more data terms tied to those products, and keep other channels ready.
Watch October 30. That is when the terms bind. Compare the live text with the preview, and watch for a fuller account of what Meta's crawlers collect.

The story

Meta rewrote its new commercial terms on Friday, October 17, deleting a sentence that legal experts said would have let it take content from business customers' websites, news publishers included, and use it to train its AI products. The Verge reported the edit a day after the clause drew criticism. Meta attached a statement saying its crawling provisions were only ever meant for its ads and business products. Spokesperson Daniel Roberts told The Verge, "This is not about scraping news sites."

The terms take effect on Friday, October 30, and apply to every company with a Meta business account.

What Meta's commercial terms said before the edit

The preview of the terms had businesses authorize Meta, or anyone working for it, to point scrapers, bots and crawlers at any domain tied to their products or services, to create test credentials, and to keep fetching fresh copies over time. A further sentence allowed Meta to use whatever it gathered to build, research, improve and train its products. There was no opt out: a business that refused could stop using Meta commercially.

Nieman Lab first flagged the clause on Thursday. Almost every newsroom runs a Meta business account, so the language reached far past online stores. Experts told the outlet Meta could cite the contract to argue it outranks a publisher's robots.txt block. Attorney Matt Topic said that access would likely be worth many millions of dollars if bought through licensing. Meta already has AI licensing deals with People Inc., CNN, USA Today and Fox News.

Why Meta called it a consolidation

Meta framed the update as folding older agreements into one. Communications chief Andy Stone wrote that the crawling language lets Meta grab accurate details about products sold on outside sites for ad and content recommendations. Ars Technica compared old and new text and found the earlier crawling permission, in Meta's generative AI ad creative terms, was confined to making a customer's own ads. Ars also found no mention of test credentials in the agreements being merged. Meta pointed instead to a separate seller agreement that uses the phrase for e-commerce listings.

Nieman Lab also pointed to Meta One, a new subscription that charges businesses and creators to put links in posts and reels. TechCrunch noted that Meta beat a major AI copyright suit last year, while other cases against it continue.

What we don't know yet

Whether the scraping and test credential authorization survives now that the training sentence is gone.
How narrowly the revised text defines ads and business products.
Whether publishers pull back from Meta business accounts before October 30.