Meta’s Updated Business Terms Still Leave Door Open for AI Scraping of News Sites

Meta recently rewrote the business‑account terms that take effect on October 30, removing explicit references to “training” its products with material taken from linked websites. However, scholars such as Cornell’s James Grimmelmann argue that the remaining clause allowing Meta to “develop” any “business product” with that content is broad enough to cover model training.

Attorney Matt Topic adds that the term “develop” is not defined in the agreement, leaving room for Meta to claim that building AI services falls under the permitted uses. The company insists the update is not intended to override robots.txt directives, yet the contractual language could still be cited in court to justify large‑scale scraping.

Why it matters for GPU / AI infrastructure

If major platforms can legally harvest publisher data without clear consent, the demand for high‑performance GPU clusters to ingest and process that data will rise. Cloud providers and hardware vendors must anticipate workloads that involve massive, potentially unlicensed web corpora, influencing capacity planning and compliance tooling.

Publishers are now evaluating technical barriers such as stricter robots.txt rules, authentication walls, and legal action to protect their assets. The outcome will shape how AI developers source training data and how infrastructure operators allocate compute resources for legitimate versus contested datasets.