Meta drops the word 'train' from commercial terms before Oct 30 rollout

By Marcus Bennett

Meta has removed the word "train" from its "streamlined" commercial terms of service, which take effect on October 30. The change follows a week of backlash triggered by legal experts who warned that the new terms appeared to grant Meta sweeping scraping rights over websites linked on business pages.

The edit is narrow. The consequence is not.

What did the original terms appear to allow?

Legal experts who spoke to Nieman Lab warned that Meta's new commercial terms of service seemed to give the company permission to endlessly scrape websites linked on business pages. The concern centered on a single mechanism: any business that operates a page on Meta's platforms would, under the terms, appear to hand over broad access to data as a condition of simply running that page.

Experts suggested a specific financial stakes behind the language. They told Nieman Lab that Meta could possibly use the policy to avoid millions in licensing fees from news organizations by requiring all business users to share a wealth of AI training data simply for running a business page.

The context matters here. News publishers have spent years negotiating licensing deals with platforms that use their content, and regulators in multiple jurisdictions have pushed platforms to pay for that use. A terms-of-service clause that obligates businesses — including publishers — to supply training data as the price of maintaining a page would shift that calculus entirely. The millions at stake, per the experts cited by Nieman Lab, are licensing fees Meta might otherwise owe.

What did Meta change, and when?

Meta dropped the word "train" from the commercial terms ahead of the October 30 effective date. The company had described the terms as a "streamlined" update when it introduced them.

A Meta spokesperson confirmed to Ars Technica that Meta did not alert users to the changed language following the backlash. That omission carries practical weight. Users comparing the announced policy against the version that now goes live on October 30 have no notice from the company explaining what moved, or why.

In response to the backlash, Meta's spokesperson provided a statement to Ars Technica. The statement acknowledged Nieman Lab's initial report but did not acknowledge the follow-up report in which Nieman Lab reiterated the experts' concerns. Meta has not, according to the available reporting, disputed the substance of what the experts described.

Why does a single word matter?

In AI policy disputes, the word "train" carries a specific technical and legal weight. Training (the process of feeding data into a model so it learns patterns from it) is the stage at which data-use questions — consent, licensing, copyright — are contested. Removing the word from the terms does not, on its own, answer the underlying question of what Meta does with data collected under the policy.

The experts' concern was not about vocabulary. It was about scope: endless scraping of websites linked from business pages, with the terms functioning as the legal hook. Dropping "train" removes one flashpoint term while leaving the broader mechanics of the policy, as described in the reporting, in place ahead of the October 30 rollout.

What happens on October 30?

The commercial terms take effect that day. Business users — a category that includes news organizations, brands, and any entity operating a professional presence on Meta platforms — fall under them as scheduled.

The open questions are concrete. Meta has not alerted users to the language change, per the company's own spokesperson. The company's statement addressed Nieman Lab's first report while leaving its follow-up — the one restating the experts' concerns — unaddressed. And the experts' central warning, that the terms could substitute for millions in licensing fees by converting business pages into a supply of AI training data, stands without a direct rebuttal in the record.

For publishers and business users, the practical move before October 30 is to read the terms as they now stand, not as they were announced. For Meta, the episode shows the cost of quiet edits: a one-word change, made without notice, now defines how the entire policy will be scrutinized.

The rollout proceeds on October 30. What Meta does with data gathered under the revised terms — and whether regulators or publishers treat the scraping concern as resolved or merely renamed — will determine whether this was a concession or a rewording.