“Does Meta think we can’t read?” Meta abandons sketchy commercial terms update.

Editor's note: When user agreements quietly become a “license” for training data, how much real choice do users have left? Meta's retreat this time looks more like a tactical withdrawal under intense public pressure than a change of position.

A Terms Update That Was “Read Out Loud”

According to Ars Technica reporter Ashley Belanger, Meta has abandoned its planned update to its commercial terms. Under the original plan, owners of Facebook pages would have been required to accept new terms whose substance was to allow Meta to use content linked to or associated with those pages for training artificial intelligence models. In other words, as long as a company or creator continued to operate its page, its content could automatically enter the training corpus.

After the news emerged, questions quickly flooded social platforms, and one widely circulated challenge was especially pointed: Does Meta think we can't read? Under public pressure, Meta ultimately chose to back down—the company will no longer force page owners to accept this authorization, and page administrators will not be deemed to have automatically agreed to their content being used for AI training simply because they continue using the platform.

The Core Dispute: Default Consent and Bundled Authorization

The key issue in this controversy is not whether Meta uses public content to train models—it has long been doing so—but how “consent” is constructed. By putting data authorization into a lengthy commercial terms document tightly bound to day-to-day operations, users who want to continue using page features have almost no choice but to accept. This design is known in the industry as “bundled consent”: there is choice in appearance, but not in practice.

For page owners with large amounts of original text, images, videos, and customer resources, it is often difficult to predict what it means for their content to be included in model training. It could mean brand exposure, but it could also mean that once the model has “learned” the content, the content's commercial value is diluted in reverse. The terms update triggered strong backlash precisely because it touched the most sensitive nerve for creators: control over their content.

Industry Context: A Data Contest Without an Endpoint

Meta's approach is not an isolated case. Since 2024, Meta has used public posts on Facebook and Instagram to train its AI models and has provided users with an opt-out channel (deeply buried); in the EU and the UK, due to regulations such as GDPR, related plans were temporarily suspended. The privacy advocacy group NOYB filed complaints in multiple European countries, forcing Meta to adjust its strategy.

Across the industry, platforms such as LinkedIn, Reddit, Stack Overflow, and Adobe have all faced controversy over similar terms adjustments or content licensing deals. Competition in model capabilities is essentially competition for high-quality corpora, and the content that users produce day after day on platforms is the cheapest and most abundant “fuel.” As a result, revising terms of service became the least troublesome way to collect it—until users began reading every word.

Analysis and Opinion: Retreat Is Not Change

In terms of results, Meta's withdrawal is a correction worthy of recognition, but it looks more like a stress response to public relations risk than an upgrade in data governance philosophy. Terms can be withdrawn, but training that has already occurred cannot be rolled back; giving up forced authorization today does not mean that it will not be repackaged in another, more covert way in the future.

The real solution may lie in three points: first, separate AI training authorization from terms of service and make it an independent, clear, and revocable choice; second, provide visible transparency so users know which of their content is used for training and what is trained; third, introduce external oversight so that “informed consent” moves from legal text into real experience. Users are not unable to understand the terms; they have simply long been assumed not to read them. This time, they read them—and said no.

This article was translated and compiled from Ars Technica.