Technology

Meta Shuts Down Instagram AI Feature That Generated Images From Other People's Photos Without Permission

Martin HollowayPublished 2w ago4 min readBased on 4 sources
Reading level
Meta Shuts Down Instagram AI Feature That Generated Images From Other People's Photos Without Permission

Meta has disabled a feature in its new Muse Image AI tool that let Instagram users generate AI pictures by tagging other people's accounts, pulling the capability just days after launch. The company updated its announcement post to confirm the feature is "no longer available." The Verge

Muse Image rolled out on July 7 as part of Meta's image-generation toolkit. One of its capabilities worked like this: you'd tag someone's public Instagram handle in your prompt, and the AI would analyze their posted photos—their face, style, overall appearance—to generate new images. The catch was that this worked on any public account, not just your own. Account owners had no notification that their content had been used, and no visibility into when or how often it happened unless they dug through their privacy settings.

Meta's initial solution was an opt-out toggle—a setting users could manually turn off to prevent their images from being used. But this left the default as "your public photos are available for AI reference," and put the burden entirely on users to find and enable the protection. SAG-AFTRA, the actors' union, flagged this immediately, telling members to opt out and publishing step-by-step instructions. For actors and public figures, whose faces are their livelihood and inherently public, this posed a particular risk.

The controversial part was the silence. Meta didn't notify anyone whose account had been tapped for image generation. The New York Times You'd only discover your likeness had been used if an AI-generated image of you popped up on the platform or elsewhere online. NBC Bay Area By July 10, days after launch, Meta had disabled the feature entirely.

The consent question at the core of this episode matters for how AI tools are built more broadly. There are two ways to handle whether someone's image can be used as AI reference material. An opt-out model assumes that if you don't actively disable it, it's fine—your image is fair game by default. An opt-in model is the reverse: unless you explicitly agree, your image stays out of the system. These are not minor technical choices. With opt-out, the discovery burden falls on users. With opt-in, the platform bears the burden of getting permission first. Meta chose opt-out at launch.

The broader context here is what this pattern tells us about how some platform companies approach consent in AI tools. Meta has collided with likeness rights issues before, iterating on image-based AI features multiple times across its platforms, each time pulling back after public pressure rather than designing the consent question in from the start. This week's shutdown of Muse's tagging feature fits that arc.

In my view, there's a product culture problem underneath this pattern. Consent keeps getting treated as a checkbox in the settings menu rather than a foundational part of how the tool works. Public accounts are visible to everyone, true—but that visibility has never meant a platform can feed that content into any AI system it builds. As generative tools proliferate, companies will need clear legal and product thinking on that distinction before launch, not after.

What Meta chose to do is also telling. The company could have reworked the feature to require explicit opt-in, added notifications so people would know when their images were used, or limited the tagging tool to your own account only. Instead, it removed the feature entirely. That suggests either that no quick fix resolved the legal or reputational risk, or that the feature itself wasn't worth preserving while the company rethinks how consent should actually work. Meta hasn't said which. The Verge

For teams building AI tools elsewhere, the lesson from this week is practical: regulators, unions, and users are drawing a clear line in 2026 around generative reference features. Public availability of an image no longer equals permission for AI reuse. And silent operation—no notification, no record visible to the person affected—counts as an aggravating factor, not a neutral detail. Meta hasn't announced whether a revised version of Muse's image-reference capability will return.