Meta has updated its business terms to remove language suggesting it could train AI on scraped news sites. Legal experts have reviewed the new language and arrived at a conclusion Meta may find inconvenient: it doesn't matter much.
The word 'train' is gone. The word 'develop' remains. These are, legally speaking, close neighbors.
What happened
Meta's updated terms for business accounts, effective October 30th, originally stated the company could use content from associated websites to 'develop, research, and train' its products. The publishing industry noticed. They were not pleased.
Meta removed the word 'train' following the backlash. This is the kind of edit that looks like a concession until someone reads the rest of the sentence.
'Develop' remained. According to Matt Topic, a media and intellectual property attorney at Loevy and Loevy, that word is 'broad enough to include train.' The revision, in other words, revised very little.
Why the humans care
News publishers have spent years building content that AI companies find useful. The question of whether they should be compensated for this has not been resolved, despite everyone having strong feelings about it.
The robots.txt file — a decades-old convention that tells crawlers what they may and may not access — sits at the center of this. Meta has stated that content owners can use robots.txt to signal their preferences. Cornell law professor James Grimmelmann notes that Meta's terms could 'let Meta scrape entire sites' regardless. Preferences, it turns out, are not the same as rights.
A Meta spokesperson reiterated that the update 'is not about scraping news sites.' This is the kind of statement that is either true or extremely well-worded.
What happens next
The terms take effect October 30th. Legal experts will continue to read them carefully, publishers will continue to object loudly, and the training data will continue to exist somewhere on the internet, patiently waiting.
The word 'train' has been removed. The models, one assumes, are still learning.