Meta's Muse AI Builds Detailed Profiles of Users' Friends, Family

What happened

Meta's new conversational AI agent, Muse, has been downloaded by millions of users, according to reporting from Wired. To make itself more useful — following instructions, suggesting actions, personalising replies — Muse builds out detailed profiles not just of the person using it, but of the friends and family members that person talks about or interacts with through the assistant.

The reporting frames this as a trade-off: the more context Muse gathers about the people in a user's life, the more capable it becomes at carrying out requests on their behalf. But that capability is built on inferring and retaining personal detail about third parties who never directly consented to being profiled by the assistant.

Why it matters

Consumer AI agents are increasingly judged on how well they anticipate intent and personalise responses, and that performance is driven by how much data the system can hold about a user's world — not just the user themselves. Muse illustrates a pattern likely to recur across the industry: assistants that feel most "helpful" are often the ones quietly accumulating the richest shadow profiles of people who are not the account holder.

For organisations building or deploying AI agents — whether for consumers, employees or customers — this raises a design and governance question that goes beyond the usual first-party privacy policy: what happens to data about people who are mentioned, referenced or implicated by someone else's use of the tool, and who is accountable for that data's lifecycle.

The Renascence take

The Muse story is less about Meta specifically and more about a structural blind spot in how "helpful" AI is engineered: usefulness and third-party data exposure tend to rise together, and most consent frameworks were never built to handle that.

Most privacy design still assumes a single, consenting user — but an AI agent that gets smarter by learning about the people around that user is quietly creating profiles nobody agreed to. The behavioral reality is that people trade away others' privacy far more readily than their own, especially when the assistant feels conversational rather than transactional. Any organisation shipping an AI agent should be asking not "did our user consent" but "whose data is this system accumulating that never had a say" — and building visible controls that let the primary user limit, inspect or delete what's been inferred about third parties, not just themselves.