Meta's Muse Image: When Your Face Becomes an AI Prompt

Public visibility is being treated as consent to turn personal identity into AI raw material, notes Mishi Choudhary.

Platform consent between Meta and its users cannot manufacture consent from everyone else who becomes part of its AI ecosystem.

(Photo: Namita Chauhan/The Quint)

Meta is on a roll these days. From introducing privacy-destroying surveillance fashion glasses to privacy-protecting usernames for Whatsapp to now furtively treating our public images as training data through Muse Image, the company has not left the news cycle.

Introduced last week, Muse Image is the first image-generation model from Meta Superintelligence Labs, integrated with Meta AI. One of its features that created a furore worldwide was Muse Image’s ability to generate AI images using photos from public Instagram accounts.

If a user’s profile was public, another user could tag that account and use their images as part of an AI generated creation. The only exception to this feature were the private accounts and accounts belonging to users under 18.

Not that the company has a stellar record on privacy, but in this particular instance, even users were to have no notification, no idea that their public photos were incorporated into AI generated images by strangers. You may have uploaded a profile picture to identify yourself—or communicate with family and friends—without realising that anyone could turn you into AI raw material and reuse your public content.

There was an opt-out feature, meaning you had to actively go through steps to opt out of it. This is deliberate friction by design.

The person whose identity was used was not necessarily asked for permission or notified of the resulting “creative experience”. Suddenly one’s decision to put a public picture was automatically being treated as if it was available for every downstream AI use. As the possibility of harassment, impersonation, and real harm to people was highlighted, Meta was forced to roll back the feature.

Indian Law Protects Celebrities. What About Everyone Else?

This may not be the last time we see this movie. As tech companies continue to integrate AI into every platform, feature, and tool without transparency and clarity about reuse of user data, we will have several versions and iterations of such features.

This story, therefore, is not about a dangerous feature that was rolled back but about product design choices that assume public visibility is a sufficient basis for treating any information as raw material for training AI models.

Indian courts are full of pending cases, as well as interim orders establishing and recognising some kind of personality rights and identity for well-known celebrities. These cases increasingly recognise a person's name, voice, and likeness, yet the practical benefit of this jurisprudence remains largely with celebrities. But these cases are about people who identify misuse, can move courts, have commercial value in their persona and can therefore receive injunctive relief. 

This completely ignores an ordinary person whose public image may be used to create sexualised or fraudulent content may not have the luxury of getting such relief. They are left to struggle to establish any such commercial rights. They will then be forced to participate in the experiment and be relegated to a separate privacy devoid class.

Ordinary folks have the same interest in their dignity and autonomy. Why should control over how we are presented or how our identities are used be the sole right of famous people? Appropriation of identity cannot be fame dependent when ordinary citizens possess only the theoretical ability to click “Accept”.

In any case, courts consider these cases once the output is already produced—be it a deepfake, a cloned voice, or a fabricated endorsement of a product. Here, the issue is the product designing a system to use another’s identity for creating training data. A much harder question for which no apparent harm may be visible today, but is ripe for future exploitation.

In principle, the Digital Personal Data Protection (DPDP) Act, 2023 and framework with respect to consent, notice, and withdrawal should be highly relevant here, but it's not fully operational yet.

Even after the full rollout, it won’t have the teeth to address the issue given it does not create rights against synthetic impersonation. Platform consent is often bundled and buried, detached from particular AI use.

The Next Frontier of AI Surveillance

Meta has also glamorised its AI glasses—and is selling it through Kylie Jenner now. These glasses combine camera, microphone, and other sensors connected to AI, allowing the person wearing them to record their surroundings. The glasses have the ability to bring such bystanders into the Meta ecosystem without them ever having an inkling about what is being captured about them. 

Platform terms between Meta and user cannot manufacture consent of bystanders. Muse Image and Meta AI glasses are part of the same governance problem. One turns existing social identity into a training material. The other turns surrounding people and places into ambient input for Meta. In both cases, the person affected may not be the person operating the tool or device allowing identity appropriation as a product feature at industrialised scale.

In a country like India, identity has different facets, constantly challenged by our society and state forcing the citizens to produce records to prove who they are. Technology has always challenged our legal systems. AI is now challenging something more fundamental about our consent, identity, and autonomy.

We have not debated whether public visibility means perpetual AI permission to be used for anything or whether we ordinary citizens can retain any meaningful agency over how our identities are used in the age of generative systems. We should not wait and be outpaced by models—yet again.

(Mishi Choudhary is a technology lawyer and founder of the Software Freedom Law Center. and policy attorney. This is an opinion piece and the views expressed are the author’s own. The Quint neither endorses nor is responsible for them.)