Meta removed explicit artificial intelligence training language from its forthcoming commercial terms after criticism from news publishers and legal experts raised concerns that the social media company could claim broad rights to collect material from outside websites. The revision, confirmed in reporting published October 17, narrows the stated purposes of Meta’s automated website access while preserving provisions that allow the company to crawl websites associated with business accounts.

The change follows scrutiny of contractual provisions scheduled to take effect October 30. Those provisions govern businesses using Meta’s commercial products, including advertising services and tools connected to Facebook and Instagram. Because many publishers and other businesses maintain accounts on Meta’s platforms, the language prompted questions about whether participation in those services could effectively authorize the collection and use of independently published content.

The controversy began with an October 15 investigation by Nieman Journalism Lab, based at Harvard University, which examined the proposed commercial terms and consulted legal specialists. The original provisions authorized Meta or entities acting on its behalf to use automated software to access and collect information from domains, web pages and URLs associated with business customers. They also explicitly identified training among the purposes for which collected material could be used.

Legal experts warned that the provision could be interpreted as permission to gather articles, images and other copyrighted material from websites operated by publishers with Meta business accounts. That interpretation raised concerns because the material might have significant commercial value in the emerging market for licensing content to artificial intelligence developers.

After the criticism, Meta revised the language. The updated terms no longer explicitly include the word “train” in the disputed provision. They also describe the relevant purposes as developing, improving, researching and maintaining advertising and business products, rather than referring broadly to Meta’s entire product portfolio. Nieman Lab reported that the revised version appeared Friday afternoon, October 16.

Meta has maintained that the update does not represent a change in its actual website-crawling practices. In a statement reported by The Verge on October 17, company spokesperson Daniel Roberts said the provisions concerned advertising and business products, not a new authorization to scrape news sites. The company also said website owners could continue expressing their crawling preferences through robots.txt files, a common technical mechanism for communicating automated-access instructions.

That distinction is central to the dispute. A revision to contractual language does not necessarily establish that any previously collected material has been deleted, that particular AI training activities have stopped or that the company’s technical systems have changed. The reported developments concern the permissions described in commercial terms, not independently verified changes to Meta’s underlying data collection operations.

Meta’s revised provisions still permit automated access to websites and material associated with commercial accounts. They describe activities including indexing, caching, analyzing, collecting and retrieving information. The language also continues to permit the creation of test credentials for certain access activities. Critics argue that these permissions remain broad even without an explicit reference to training AI models.

James Grimmelmann, a professor of digital and information law at Cornell Law School, told Nieman Lab that the revisions did not resolve his central objections. He argued that the language concerning the development of advertising and business products remained expansive and that the provision connecting website access to a business customer’s products or services was not sufficiently limited.

Matt Topic, a media and intellectual property attorney at Loevy & Loevy, similarly questioned whether deleting the explicit training reference meaningfully narrowed the provision. He noted that AI systems can be part of business product development, potentially leaving room for contested interpretations of how collected website content may be used.

The legal distinction matters because artificial intelligence training is not the only commercially valuable use of online information. Automated collection can support search functions, advertising systems, business analytics, recommendation technologies and other software development activities. The revised terms therefore raise questions about the breadth of permissible commercial uses even if Meta’s explanation of its intentions is accepted.

Another unresolved issue involves robots.txt, a file that website operators use to communicate instructions to automated crawlers. Meta says publishers and other content owners can use the mechanism to indicate their preferences and that the revised terms do not override those instructions. Legal experts, however, distinguish technical instructions from enforceable contractual restrictions. A website can publish crawler preferences without necessarily establishing an absolute legal prohibition on every form of automated access.

Benjamin Sobel, a law professor at the University of Wisconsin, told Nieman Lab that compliance with robots.txt preferences is generally voluntary. His analysis highlights the potential gap between a company’s public assurances about honoring website restrictions and the rights that might be asserted under a broad commercial agreement. The precise relationship between those mechanisms could become important if a dispute arises over access to copyrighted material.

The backlash has extended beyond individual website operators. On October 16, the News/Media Alliance, a U.S. trade association representing news and magazine publishers, issued a statement sharply criticizing Meta’s proposed provisions. President and CEO Danielle Coffey argued that conditioning access to Meta’s commercial audience on acceptance of sweeping website-collection permissions created an unacceptable imbalance between the platform and content owners.

The organization’s concerns extend to the economics of digital publishing. News organizations invest in reporting, photography, editorial production and distribution while relying on social networks to reach audiences. If a platform can interpret commercial account agreements as granting substantial rights over material hosted elsewhere, publishers may face additional uncertainty about how that work is monetized or incorporated into competing technology products.

Similar concerns could affect businesses outside the news industry. Retailers, manufacturers, entertainment companies and independent creators may operate websites containing proprietary descriptions, images, technical information or other valuable material while maintaining Meta accounts for marketing purposes. The question is whether using an advertising or business-management service should also create permissions over content that exists independently of that service.

The upcoming implementation date adds urgency. Nieman Lab reported that businesses did not have a separate option to reject the new commercial terms while continuing to use the covered Meta services. That arrangement potentially leaves customers choosing between accepting the revised agreement and discontinuing commercial use of the affected products. The practical implications will vary according to each company’s dependence on Meta’s advertising and audience-distribution infrastructure.

For Silicon Valley, the dispute illustrates a broader commercial tension surrounding artificial intelligence development. Technology companies increasingly depend on large quantities of external information to build and improve sophisticated products. Publishers and other rights holders, meanwhile, are seeking greater control over the conditions under which their material is collected, licensed and reused.

Meta’s position is that its contractual changes clarify existing permissions rather than expand them. Critics counter that consolidating provisions previously associated with more specific advertising features into broader commercial terms could materially affect the scope of authorization. Nieman Lab reported that the previous crawling language was tied to generative AI advertising features, while the updated terms cover a wider range of advertising and business products.

The distinction has not yet been settled by a court in relation to these particular terms. Nor does the October 17 revision establish that Meta has infringed any publisher’s copyright or improperly trained an AI model using material obtained under the disputed provision. Those conclusions would require evidence about actual collection and use, alongside applicable legal analysis.

For now, the confirmed development is a rapid modification of contractual language following public scrutiny. Meta has removed the clearest reference to AI training and more explicitly limited the stated purposes of automated collection. The company continues to deny that the change alters its crawling practices or overrides website owners’ expressed preferences.

With the revised commercial terms scheduled to take effect October 30, the remaining questions concern the reach of the crawling authorization, the meaning of business product development and the effectiveness of existing controls available to content owners. The outcome could influence how American technology platforms describe data-access permissions and how businesses evaluate the relationship between their digital distribution channels and intellectual property.