The announcement came in a post on X from Meta Platforms Inc.’s communications chief, Andy Stone, rather than in a filing. He said the company does “not plan to make policy changes without a full consultation with industry groups and leaders.”
The practical effect, confirmed by the company on Oct. 28, is that the new commercial terms will not take effect on Nov. 20 as scheduled, and may not take effect for some time. It is the second postponement. The terms were originally set to apply on Oct. 30.
The dispute is over what a business account on Meta’s platforms agrees to when it accepts the company’s standard terms. As drafted, the language covers websites “associated with” accounts on Facebook, Instagram, WhatsApp and Threads, and has been read by publishers as permission for Meta to use their pages, images and posts to train its artificial intelligence models.
Publishers in Europe and Britain objected first, and then digital advertising firms joined them. On Oct. 22, trade bodies in several countries sent letters demanding that Meta withdraw the terms. The News/Media Alliance, which represents more than 2,000 publishers, wrote to Chief Executive Mark Zuckerberg asking him to scrap the revision before the original deadline, arguing that the choice it creates is between letting Meta use journalism without payment and losing access to the audience that lives on its apps.
Two U.S. senators added their names to the pressure. Marsha Blackburn, a Tennessee Republican who chairs the Senate Judiciary subcommittee on privacy and technology, and Adam Schiff, a California Democrat, each wrote to Meta asking it to reverse course. Both argued that the terms would let the company take the work of creators and small businesses without paying for it.
Meta has made two adjustments. It removed the word “train” from the terms on Oct. 16, after the Nieman Journalism Lab reported on the language, and it narrowed the stated use of scraped material to advertising and business products. A spokesperson said the update consolidates overlapping advertiser agreements and does not change how Meta crawls the web. Content owners, the company notes, can still use robots.txt to signal preferences, and advertisers keep their controls over generative AI features inside Ads Manager.
Publishers have said those changes answer a different question than the one they asked. Removing a word does not remove the capability, and the narrowed purpose still describes a commercial use of their material in the products Meta sells to the same advertisers.
The legal backdrop explains why the company keeps moving the date. In Europe, text and data mining is permitted for commercial purposes unless the rights holder has opted out, a reservation that has to be expressed in machine-readable form. In Britain, a similar exception carries its own conditions.
In the U.S., Meta has argued in court that training on published material is fair use, a position a federal judge in California largely accepted in 2025 while faulting the company over its use of pirated books. None of those forums has settled what a contract clause can accomplish that a copyright defense cannot.
There is also the matter of the advertisers themselves, who buy through the same agreement. Media buyers are asking whether creative work uploaded to run a campaign could end up in a training set, and what indemnity they hold if it does. Meta’s answer, that ad controls in Ads Manager govern the use of creative, has not satisfied several agency groups.
The market Meta would be sidestepping is real and priced. Rival AI developers have signed licensing agreements with large news organizations over the past two years, paying for archives and wire copy that Meta proposes to take under a terms-of-service update. Publishers point to those deals as evidence that a price exists, and that the clause is an attempt to avoid it.
For Meta, the calculation is unusually awkward. The company has spent billions on computing for AI and needs text and images to justify the spending. It also depends on publishers and creators for the content that keeps its apps worth opening, and on advertisers for essentially all of its revenue. A clause that asserts a right it may not need, in language broad enough to alarm the people who supply both, buys very little.
The company says it will keep talking. Its spokesperson said the consultation could extend beyond Nov. 20, and that implementation would be delayed again if it does. The date, in other words, is not a deadline.
What happens next depends on two rooms Meta does not control. European regulators have been examining how the company handles opt-outs from AI training, and any enforcement action would arrive with its own timetable. The copyright suits already filed against Meta will produce rulings that define what a contract can and cannot authorize.
Until then the terms sit unapplied, a document that has been amended twice, postponed twice and defended by a company that has not said what it will do if publishers hold their line. The next signal will be whether Meta publishes revised language, or simply lets the date pass in silence.
