Meta's Muse AI Creates Hidden Profiles of Friends, Family

What happened

Meta's conversational AI companion, Muse, has been found to construct detailed behind-the-scenes profiles of a user's friends and family members in order to personalise its responses, according to reporting by Wired. The feature, which operates without directly informing or seeking consent from the people being profiled, has drawn scrutiny despite Muse's rapid uptake, with the app downloaded by millions since launch.

According to the report, Muse builds these profiles by drawing on information shared during conversations — details about relationships, preferences and personal circumstances — and uses that data to shape how it responds to the primary user. The people described in these profiles are not Muse users themselves and have no visibility into, or control over, the information being compiled about them.

Why it matters

The story illustrates a widening gap between what conversational AI systems are technically capable of and what users — and the third parties swept up in their data — reasonably expect. Personalisation engines like Muse are designed to feel more useful and human by remembering context, but when that context includes inferences about people who never opted in, the product crosses from convenience into a consent and data-governance problem.

For organisations building or deploying AI agents, this is a live case study in how personalisation features can outpace privacy design. It raises concrete questions for product and legal teams: where does "remembering what you told me" end and "profiling someone else" begin, and who is accountable when an AI system makes inferences about non-users who never interacted with the platform at all.

The Renascence take

Most coverage of this story will frame it as a privacy failure. The more useful read is a service-design failure: Muse's architects optimised for one experience metric — conversational relevance — without modelling the experience of the people being talked about, who are stakeholders in the interaction even though they never touched the product.

Personalisation is a behavioral contract, not just a technical feature — it only works when the person being modelled knows the model exists. Any AI system that infers details about people outside its direct user base is making a design choice to treat convenience as more important than consent, and that choice will eventually surface as a trust cost. Operators building conversational AI should treat "who else is in this data" as a first-order design question, not an afterthought handled by the privacy team after launch.