Meta’s AI could still scrape news sites, legal experts say

Meta’s AI and News Scraping: What the Revised Terms Really Mean

Meta has recently updated its business terms, removing explicit language that suggested it could freely train its AI models on content from news sites. The change was prompted by a backlash from publishers who feared their articles would be harvested without compensation. However, legal scholars argue that the new wording still leaves ample room for Meta to scrape and use news content for AI development.

The Language Shift: From "Train" to "Develop"

The original terms, set to take effect on October 30, 2026, allowed Meta to "use any content" associated with a business account for "develop, research, and train" its products. Critics pointed out that this could override standard web‑crawling controls such as robots.txt, which publishers use to block unwanted scraping.

In response, Meta removed the word "train" from the clause, replacing it with broader phrasing like "develop any business product." While the explicit reference to training is gone, experts note that "develop" is a catch‑all term that can encompass training AI models. As media‑law attorney Matt Topic explains, the lack of a precise definition means Meta could still argue that AI model creation falls under the umbrella of product development.

Why the Change May Not Satisfy Publishers

Publishers are concerned that the revised language does not materially limit Meta’s ability to harvest large swaths of news content. The key issues include:

• Ambiguity of "business product": The term is not clearly defined, leaving room for interpretation that includes AI services.
• Broad scope of "develop": Legal analysts argue that development activities inherently involve training data, meaning the original concern remains.
• Reliance on robots.txt: Meta’s statement that publishers can still use robots.txt to signal preferences does not guarantee compliance, especially if the company claims a contractual right to access the data.

Potential Legal and Policy Implications

If Meta proceeds with large‑scale scraping under the new terms, several legal challenges could arise:

1. Copyright infringement: Using copyrighted news articles to train AI without a license may violate U.S. and international copyright law.
2. Contractual conflicts: Publishers could argue that Meta’s terms conflict with existing agreements that prohibit unauthorized data use.
3. Regulatory scrutiny: Ongoing antitrust and AI‑policy investigations may view aggressive data collection as an anti‑competitive practice.

Law professors like James Grimmelmann warn that the language “associated with your products or services” could still be interpreted to give Meta a blanket right to scrape entire sites, potentially sidestepping traditional web‑crawling norms.

What Publishers Can Do Now

While the debate continues, news organizations can take practical steps to protect their content:

• Update robots.txt: Clearly disallow all crawlers from accessing full‑text articles and specify the user‑agents you wish to block.
• Publish clear terms of use: Explicitly forbid the use of your content for AI training without a written license.
• Monitor traffic: Use analytics tools to detect unusual scraping patterns and issue cease‑and‑desist notices when necessary.
• Engage in collective action: Join industry groups lobbying for clearer legal standards on AI training data.

Bottom Line

Meta’s revised business terms may appear to address publisher concerns on the surface, but the underlying language still grants the company broad discretion to use scraped news content for AI development. Legal experts caution that without more precise definitions and enforceable restrictions, the risk of unlicensed data harvesting remains high. Publishers should stay vigilant, reinforce technical safeguards, and push for clearer regulatory guidance to ensure their content is not exploited without fair compensation.