Meta's reversal this week is a small victory for publishers, and it is also a warning. On 15 October 2026, Nieman Lab reported that Meta's updated commercial terms, due to take effect on 30 October, would let the company crawl websites linked to business accounts and use the content with no opt-out. Meta first deleted the word "train" from the clause, and legal experts said the revised wording still left room for broad scraping. It then dropped the provision altogether, according to MediaNama's summary of The Verge's reporting. The European Publishers Council said it would seek scrutiny from European regulators, and the UK News Media Association called for government intervention.
What the clause did
The disputed text said Meta "may use any content from such domains, web pages, or URLs" to provide, develop, improve, research and maintain its products. Per MediaNama, the original version also included the word "train". Business accounts had no opt-out. A business that objected was told to stop using Meta's commercial products. The terms also let Meta create "test credentials" to access sites. Cornell's James Grimmelmann said this could let the company register as a fake user to get past registration barriers.
Meta's position is that the crawling terms "only applied to our ads and business products". It says the change "does not change how Meta uses web crawlers" and that site owners can still use robots.txt. That is a coherent defence. Platforms legitimately need to fetch a linked page to render an ad preview, check a landing page or run brand-safety checks, and a consolidated terms document is a reasonable place to say so.
The strongest case for Meta's approach
The best argument for the provision is that commercial relationships can create consent that a generic crawl cannot. If an advertiser links its domain to a Meta business account, one could say it has accepted a bilateral deal on how that content may be used. Contract is also more precise than copyright law, which remains unsettled on AI training. In Kadrey v. Meta, Judge Chhabria granted Meta summary judgment on fair use on 25 June 2025. As Authors Alliance's analysis describes, the plaintiffs failed on the record before the court, while the judge floated "market dilution" as a theory future plaintiffs might use. A company facing that uncertainty may prefer explicit contractual permission to a litigated gray zone.
Why the withdrawal was still right
The problem is how the permission was obtained. A publisher that links its website to a Facebook page to reach readers is not negotiating a data licence. When the alternative to "consent" is abandoning the business tools it already depends on, the arrangement is closer to a take-it-or-leave-it condition than to a bargain. That is the contract-fairness and market-power concern the European Publishers Council says it will raise.
The design also clashed with the open web's existing preference mechanism. Robots.txt, standardised in RFC 9309 in September 2022, is explicit that "these rules are not a form of access authorization." Robots.txt is a signal and not a legal lock. Benjamin Sobel of the University of Wisconsin made a similar point to Nieman Lab, according to MediaNama's report. The signal is nonetheless the one thing the internet has for communicating preferences at scale. If a contractual term can quietly override it for any domain tied to a business account, every publisher's preferences become conditional on who it advertises with.
The regulatory baseline is already moving
In Europe the shape of the answer is visible. Article 4 of the 2019 Copyright in the Digital Single Market Directive permits text and data mining for commercial purposes unless rightholders reserve their rights in an appropriate manner, such as machine-readable means. The EU's General-Purpose AI Code of Practice, published on 10 July 2025, has a copyright chapter that the Commission describes as offering practical solutions for meeting the AI Act's obligation to comply with EU copyright law. Those rules turn on respecting rights reservations, and a hidden contractual override would sit uneasily with them. How any regulator would treat the withdrawn clause is untested, since it never took effect. What the episode shows is that the preference-signalling system is only as strong as the weakest contract that can bypass it.
Proportionate rules, not a licensing mandate
The right response is not a compulsory licensing regime or a ban on AI training from public web content. Those would raise costs for open-source developers and smaller labs far more than for Meta, and would lock in incumbents who can afford licensing deals. Three narrower measures would do the work:
- Unbundled consent. Permission to use content for AI development should be a separate, specific opt-in. It should not sit in the terms of a general advertising product.
- Honouring machine-readable signals. Contractual terms should not silently override a robots.txt or equivalent reservation, and any intended override should be disclosed prominently.
- Meaningful notice. The 15-day gap between Nieman Lab's report and the planned effective date was not a consultation. Material changes to data-use terms deserve a notice period long enough for affected parties to respond.
These are light-touch, enforceable and neutral between large and small developers. They keep the training pipeline open while ensuring that the people who publish the content can say no, or name a price.
What to watch
The clause is gone, but the commercial incentive behind it is not. Meta's spokesperson said the change does not alter how it uses web crawlers, and the company has not said whether a narrower version will return. The News/Media Alliance, per MediaNama, called the original terms "outrageous and probably illegal". If the European Publishers Council follows through, the useful outcome would be regulators stating plainly how contractual terms interact with rights reservations under EU law. That clarity would serve publishers and AI developers better than another round of unannounced edits.