I asked Muse a simple question and it returned a dossier about my sister before I could finish typing. I felt the room shift—what I thought was private felt suddenly public. You sit there and realize the assistant you welcome into your apps is quietly building a map of everyone you know.
I’ve spent years tracing how services collect the small threads that become a full picture. I’ll walk you through what I found, what researchers exposed, and what it means when an AI remembers more than you expected.
My calendar showed my sister’s birthday before I told Muse anything — how Muse builds relationship pages
Researchers found an hourly process that instructs Muse to create and update “a page for every person in the user’s life,” according to Wired and files shared by The Future Society.
Karan Joshi, an independent AI safety researcher, pulled system prompts and internal instructions from Muse’s chat interface and shared them with reporters. Those documents describe pages that collect a person’s name, how they’re connected to you, recurring topics (like a pending move or savings goal), and important dates such as birthdays and anniversaries.
It behaves like a secretary who never shuts off.
Does Muse create profiles for people who don’t use the app?
Yes. The instructions and exported virtual machine data show Muse builds profiles even for non‑users based on public details and what you share through your accounts. The Future Society published these system instructions, and Wired reported on Joshi’s extracted prompts, both confirming the behavior.
My friend’s invoice appeared with context — what Muse links together
Muse doesn’t just store names and dates. The files describe keeping a history of the relationship: events, milestones, how close you are, what the relationship is built on, and what it appears to need at the moment.
Meta told Gizmodo that Muse gathers this context from public information and from what you’ve chosen to share, giving the example of remembering that the person who sent an invoice was the plumber who fixed your bathroom or which flowers your spouse prefers.
How does Muse store relationship data?
Meta says the information lives inside each user’s dedicated secure virtual machine and cannot be accessed by other users’ agents. Independent researchers counter that the system prompts and instructions are accessible through the app’s interfaces and that any stored data becomes another attack surface—especially once an agent can act on behalf of a person.
The app shot to the top of the App Store — adoption turns theoretical risks into practical ones
Muse launched on Sept. 8 and quickly rose to the number-one spot on Apple’s U.S. App Store for free iPhone apps. With fast adoption comes a flood of granted permissions: email access, calendar entries, shopping accounts, and other connected services.
Amazon blocked Muse in September, saying Meta hadn’t told Amazon that an automated agent would access its shop and raising flags about how customer credentials and account data are handled. Meta also patched a zero-day in the Mac version that could let malware hijack an already-running Muse instance and abuse granted permissions.
I tested controls and my team asked how to stop it — practical privacy moves
You can revoke Muse’s access to apps, remove connected accounts, and audit permissions through your device and service settings. Those actions limit what Muse can ingest, but the existence of automatically generated relationship pages means traces may already be assembled.
Muse is a library where every book writes itself.
Can I delete or stop Muse from making these pages?
Short answer: partly. Remove app permissions and disconnect services to prevent new data from flowing in. Meta claims data stays in your secure VM, but researchers warn that prompts and internal instructions exposed through the app can reveal how those profiles are formed, which weakens the practical privacy guarantee.
Wired, The Future Society, Karan Joshi, Gizmodo, Apple’s App Store, Amazon, Mac security bulletins, and Meta’s own statements are the documents and brands you should watch if you care about how an agent models your social world. The risk isn’t a single bug; it’s an architecture that treats your relationships as fuel for assistance.
If Muse is building a private index of everyone you know, how comfortable are you letting an AI rewrite the margins of your life?