Meta withdrew Muse Image’s Instagram-reference feature after three days because it had built an AI product around an unusually expansive interpretation of public content. If an Instagram account was public, another user could tag it in a Meta AI prompt and generate an image based on that person’s posts. The account holder was automatically included, not asked first, and was not necessarily notified when their content was used (BBC News, 2026).
For large AI platforms, publicly accessible material has become a tempting category: cheap, abundant, varied and already tied to the kinds of human details that make synthetic images convincing. A public Instagram profile can provide faces, angles, clothing, interiors, travel, relationships, a recognizable visual style. It is a ready-made model of a person’s life.
Meta called Muse Image a creative tool that could generate personalized visuals from prompts, sketches, and reference images, and the company positioned it as part of a broader push to bring generative AI into Instagram, WhatsApp, and eventually advertising products. Yet the controversial element was not simply that the system could edit photographs. It was that it allowed a user to pull another person’s account into the process by typing an @-handle (Meta, 2026).
That mechanism removes the social friction that usually comes before using somebody else’s likeness. If you wanted to use an acquaintance’s photo in a campaign, an artwork or even a joke, you would normally have to ask. At minimum, you would have to obtain the image. Muse Image turned that into a platform action. The human being became searchable input.
Meta’s response was that people retained control because they could opt out. But this is one of those arguments that sounds reasonable until you look at the sequence required. A user must first hear that the feature exists, understand that “public content” includes their own photos, locate the setting, assess its consequences and change it. Until then, the platform has interpreted silence as permission.
That is not how consent works in most serious settings. Medical consent, for example, is specific and informed because the consequences of a decision may not be obvious. Data protection law makes a similar distinction: a person should know what they are agreeing to and have a genuine choice. Social-media companies often substitute interface convenience for that standard. A toggle buried in settings is presented as autonomy; in practice, it is frequently a test of whether users have time, technical confidence and the luck to see the right headline.
Wired reported that people whose accounts were referenced would not automatically be notified, while AI images already generated from their content could remain even if they later changed their preferences (WIRED, 2026). There is an awkward asymmetry: the person making the image gets speed and creative freedom; the person depicted receives uncertainty and a cleanup problem.
SAG-AFTRA’s objection was therefore more than celebrity anxiety. The union warned that, given the risks around non-consensual digital replicas, anything less than a clear opt-in was unacceptable. Meta removed the feature after the backlash, describing it as a misjudgement (The Guardian, 2026).
The episode also arrived as the European Commission announced preliminary findings that Facebook and Instagram’s addictive design features breached the Digital Services Act (European Commission, 2026). Again, the subjects differ, but the pattern does not. In both cases, the important decisions sit in product architecture: what is switched on, what is hidden and who bears the burden of resisting it.
The “public” is being stretched from “people can see this” to “the platform may make things from this.” That shift deserves far more scrutiny than a hurried opt-out button.