Meta drops the word "train" from its commercial TOS after AI backlash

Meta has revised the commercial terms of service that take effect October 30 after legal experts and publishers warned they let the company scrape any website linked from a business page. The word "train" is gone, and so is the phrase "any content." The new language lets Meta periodically pull updates from business websites to provide, develop, improve, research and maintain advertising and business products.

A Meta spokesperson told Ars the crawling terms only apply to ads and business products, that nothing changes about how Meta uses crawlers, and that content owners can still use robots.txt. The spokesperson also confirmed Meta did not alert users to the change. For millions of businesses, including news outlets, the condition for having a page on the world's biggest social platforms is the real issue here.

We don't think this edit fixes the problem, and the lawyers quoted agree. Removing "train" while keeping "develop" is a cosmetic change if Meta is building AI business products, which it plainly is. One Cornell law professor called the remaining language broad enough to drive a truck through. The simplest fix is the one Meta keeps declining: write into the enforceable contract that it will respect robots.txt. Saying robots.txt still works in a press statement, while leaving it out of the binding terms, is exactly backwards.

The coercion is the part that bothers us most. This is not a crawler knocking on a public site. It is a condition attached to access to Meta's audience. A local business or a newsroom that needs a Facebook page to reach people is being asked to grant content rights to its whole website as the price of entry. Trade groups in the US, UK and Europe are calling it a land grab and an abuse of market power, and the News/Media Alliance says it is probably illegal. Whether that legal theory holds is uncertain, but the leverage is real.

We want AI systems to learn from the open web; that is part of how they become useful. But the bargain has to be consent and, where the value is high, payment. Burying rights grants in platform terms and changing them quietly without notice erodes the trust that any workable licensing market depends on. If this becomes the norm, every platform with a captive business audience will try the same move.

I'm watching October 30. If regulators in Brussels or London act before then, Meta may have to go further. If not, this language becomes the template.

Nieman Lab first reported the concerns, and its follow-up said the revised terms still left publishers exposed. The European Publishers Council has asked the European Commission to intervene, and the UK's News Media Association has urged British regulators to act. Compliance with robots.txt is voluntary, which is why experts want it written into the contract.