Meta Removes 'Train' from AI Scraping Terms After Backlash, But Critics Remain Unconvinced

Image: Ars Technica · Source

Meta revises its October 30 commercial terms of service, dropping the word 'train' related to AI data use, yet experts and publishers argue the changes still enable broad website content scraping without sufficient user protections.

Meta recently modified its upcoming commercial terms of service (TOS), which are set to take effect on October 30, 2026, following significant criticism concerning the company’s approach to AI training data scraping. The key update involves the removal of the word "train" from the TOS language, which initially allowed Meta to use content linked on business pages to "train" its AI and other products.

The original terms appeared to grant Meta broad rights to scrape "any content" from websites connected to business pages on its platform, with wording that permitted using this data to "provide, develop, improve, research, train, and maintain" Meta's products and services. This sparked alarm among legal experts and media groups, fearing Meta was exploiting business users and circumventing licensing fees by harvesting vast amounts of web content for AI development without proper consent.

Meta’s spokesperson told Ars Technica and Nieman Lab that the crawling terms specifically apply to Meta’s advertising and business products and do not grant new rights superseding existing content owner preferences, including those indicated via robots.txt. Despite the clarification, Meta did not actively notify users of these revisions prior to the policy rollout, adding to public confusion.

The updated terms now replace the phrase "any content" with allowances for Meta to "periodically pull updates" from business users’ websites to support "advertising and business products" and to "maintain the integrity and security" of Meta’s offerings. The company claims this scraping serves to benefit users by enhancing Meta's understanding of products and helping connect businesses with prospective customers.

Nevertheless, critics remain unconvinced the amendment sufficiently protects content owners. Matt Topic, an attorney specializing in media and intellectual property, emphasized that permits to "develop any 'business product'" remain overly broad and could still encompass AI training activities. Cornell Law School professor James Grimmelmann echoed these concerns, noting that the vague language leaves loopholes for expansive scraping.

A significant shortfall noted by experts is the absence of explicit commitments from Meta to respect website owners' robots.txt directives. Compliance with robots.txt is voluntary, and without explicit TOS language enforcing these preferences, Meta could potentially bypass site-level restrictions, intensifying publisher anxieties.

The backlash against Meta's TOS extends internationally. UK News Media Association CEO Theo Bamber condemned the terms as an "arrogant tech platform bullying" attempt to seize content without fair negotiation. The European Publishers Council and US-based News/Media Alliance have called on their respective regulators and governments to intervene, with concerns focusing on both the legal and ethical implications of Meta’s sweeping content use policies.

Danielle Coffey, president and CEO of the News/Media Alliance, described Meta’s tactics as "an appalling, bad-faith policy" that abuses its market power by effectively forcing content owners to allow scraping to maintain platform access. These groups urge regulators to block or substantially revise the upcoming TOS before launch.

Meta’s revisions demonstrate a response to public and legal pressure, but industry experts and stakeholders remain wary that the fundamental issues of consent, fair compensation, and respect for website content remain unresolved. How regulators and Meta will navigate these concerns ahead of the October 30 enforcement remains to be seen.