Meta AI Generates Images Using Instagram Users' Likeness Without Consent

Meta has rolled out a new generative model called Muse Image inside the Instagram app, and it is switched on by default across millions of public profiles. Type someone else's Instagram handle into a prompt, and the algorithm pulls from that person's public photo gallery to generate new, photorealistic images, altering the background, outfit, or context of the scene.

The mechanism is described by forsal.pl, which notes that generating an image with someone else's likeness requires no downloaded photos and no access beyond what that person has already posted publicly. The account handle alone is enough for the system to automatically search the gallery and use it as reference material for a new composition.

How the generation bypasses consent

In practice, this means any user with a public profile becomes potential source material for someone else's prompts. The model can change the setting, clothing, and situation a person appears in, opening the door to sophisticated image manipulation bordering on deepfakes, created without the knowledge of the person pictured.

The account owner receives no notification that their likeness was used to generate a new image. There is also no mechanism to check how many times, or by whom, their photos were used as reference material for the algorithm.

A hidden off switch

To block Muse Image from using your own photos, you have to open profile settings, then the section on interacting with other users, find the subsection on sharing and reusing content, and finally flip separate toggles for posts and for Reels under the Meta AI tools section. The path is far from obvious, and since the feature is on by default, user inaction amounts to automatic consent.

Even after the feature is switched off, content generated earlier stays in circulation. There is no option to retroactively remove images already created from someone's likeness before they managed to lock down their profile.

A gap involving children's photos

Meta says profiles belonging to minors are excluded from the model. The problem is that many photos of children reach the platform through their parents' or guardians' accounts, which belong to adults. Since the adult's account, not the child's, is the formal owner of the post, those photographs remain within the algorithm's reach despite the stated safeguard.

This practice, known as sharenting, parents posting photos of their children online, has long drawn criticism from data protection specialists. Introducing a generative model that can process such photos without additional age verification adds a new layer of risk to an already known problem.

What it means for Polish users

Instagram has millions of active accounts in Poland, a large share of which remain public, especially among younger users and creators. Because Muse Image is on by default, Polish users are already covered by the feature unless they have manually turned it off, and most likely do not even know it exists.

The case could also draw the attention of European data protection regulators, since a mechanism that processes someone's likeness by default, without explicit consent, raises questions under EU personal data protection rules and the incoming AI Act. Meta has not announced any plan to change the default settings to a model requiring informed consent.

Sources: New Meta AI model hits user privacy (forsal.pl)