I'm Nitin Kumar, Lead Project Coordinator at Debugged Pro. I help software and IT agencies in India and the USA brief clients on AI agent privacy before pilots go live. This week WIRED reported that Muse relationship profiles can build a detailed page for people in a user's life. That changes how you talk about Meta Muse privacy on delivery calls.

Millions have downloaded Muse. Many users connect bank accounts, messages, or health data so the agent can finish tasks. Researchers then extracted Muse instructions through the normal chat interface. Meta said those files were meant to be accessible for transparency.

This guide sticks to verified facts from the WIRED Kernel Panic report on Muse relationship profiles and corroborating notes from the Future Society Muse findings review. It is a privacy and memory angle, not a redo of Muse gadgets or small business checklists.

Muse relationship profiles and Meta Muse privacy illustration from WIRED coverage — Feature image based on WIRED photo-illustration for Muse Kernel Panic coverage. Credit: WIRED Staff / Getty Images via media.wired.com.

What Muse relationship profiles actually do

Independent researcher Karan Joshi extracted Muse instructions and system prompts. He shared findings with WIRED. One instruction describes creating a page for every person in the user's life.

The process runs on an hourly cadence. It covers family, partners, friends, colleagues, collaborators, and people you follow. Muse memory here means structured text files, not a vague chat summary.

Joshi told WIRED it seemed Meta wants to understand relationships with real people. He said they are trying to know you like a friend, and called that creepy. Agencies should quote that as a user trust signal, not as legal advice.

Muse people pages and section labels

Documentation says a page may start sparse and fill out over time. Reported sections include Facts, History, The relationship, In common, Open threads, and Strengthening. Instructions say Muse should use only available evidence.

Facts: where people live, what they do, dates that matter such as birthdays or anniversaries.
History: backstory like a trip in March, an argument that got resolved, or a recent milestone.
The relationship: how close they are, what it is built on, how they act, and what it seems to need.
In common / Open threads: recurring topics such as an apartment move or a shared savings goal.
Strengthening: a reason to call, a date worth remembering, something to circle back on, or a way to be there.

Invented details are framed as worse than an empty page. That rule matters when clients ask if the agent will hallucinate contacts. Still treat every stored inference as sensitive AI agent relationship data.

Meta Muse privacy controls including wipe memories, disconnect services, and audit log — Debugged Pro illustrative card summarizing Meta Muse privacy controls reported by WIRED. Not an official Meta UI screenshot.

Meta Muse privacy architecture clients ask about

WIRED summarizes the product architecture Meta describes for consumer Muse. Each user has a dedicated virtual machine that stores data and context. That VM is inaccessible to other agents.

Users can wipe memories or disconnect external services at any time.
Muse seeks human confirmation before sending an email or making a purchase.
An audit log shows agent activity and future plans.
Meta spokesperson Daniel Roberts said Muse gathers context from public information and what the user chose to share.

Roberts gave examples such as linking an invoice sender to a plumber hired earlier, or remembering flowers a spouse liked. Those examples sound helpful in demos. They also show how fast ordinary connectors become relationship graphs.

What Future Society adds on Muse memory

The Future Society review titled Muse, in its own files is based on a September 24, 2026 VM export. Developers prompted Muse to archive nearly its entire filesystem. Meta called that intended behavior because each VM is isolated to its own user.

One high-severity finding matches the WIRED people-page story. An hourly job writes a page about each person with a real tie to the user. That can include people who never signed up for Muse or agreed to it.

Another finding says asking Muse to forget something edits memory notes but keeps the chat transcript. The forget flow is told not to mention that the chat remains. Put that in your Meta Muse privacy FAQ for client stakeholders.

Future Society also notes memory embeddings, fleet-learning outboxes, and lessons published centrally outside the per-user VM. Use that only as a research citation. Confirm current product claims with Meta before you put them in a signed SOW.

Muse people pages sections Facts History Relationship In common Open threads Strengthening — Debugged Pro illustrative summary of Muse people page sections described in WIRED reporting. For agency briefing use only.

Why Muse relationship profiles raise asymmetric risk

Carissa Veliz, associate professor at Oxford's Institute for Ethics in AI, told WIRED users give AI systems more information than they get back. Her point covers explicit sharing and inferences, correct or incorrect. See the Oxford Institute for Ethics in AI for her institutional home page context.

Miranda Bogen, director of the Center for Democracy and Technology AI Governance Lab, said Muse appears to emphasize relationships more than rival systems. She also noted assistants solicit email, calendar, and financial access. That breadth balloons what tools know about users.

For India and USA agency delivery teams, the practical issue is third-party data. A client's Muse can store notes about employees, vendors, and family contacts who never clicked accept. Your privacy review must cover those people, not only the account holder.

Delivery checklist for India and USA agency teams

Use this checklist on discovery calls before anyone wires production calendars into Muse. Keep consumer experiments off shared agency laptops when possible. When Mac desktop agents are also in scope, review Full Disk Access controls in a separate privacy pass.

Map connectors. List banks, messages, health, email, and calendar access the pilot needs. Refuse unused connectors.
Name relationship data. Say aloud that Muse relationship profiles may store Facts, History, and Strengthening notes about contacts.
Run a wipe drill. Have the client wipe memories, disconnect a service, and confirm audit-log entries update.
Test forget vs transcript. Ask whether forgotten notes still appear in chat history after a forget request.
Block production secrets. Ban passwords, OTPs, card numbers, and health measurements from agent memory prompts.
Separate hardware paths. Do not mix Muse gadget DIY projects with privacy pilots. Keep hardware experiments on a different brief.
Write SOW language. State that third parties may be profiled without their signup. Assign who reviews the audit log weekly.
Keep delivery systems separate. Client Muse accounts are not your agency CRM, ticketing system, or code host.

Questions to ask on the first privacy call

Which roles may connect Muse to company email or finance tools?
Who owns the wipe and disconnect process when staff leave?
How will you notify people who appear on Muse people pages?
What happens if Muse suggests relationship actions that touch HR or legal issues?
Where does AI agent relationship data sit relative to GDPR, CCPA, or India DPDP obligations?

If the client needs implementation help, start with generative AI consulting for agent design or software development consulting for integration and governance work. Earlier Muse small-business checklists cover ops use cases, not this people-memory risk.

How Muse memory differs from ordinary chatbot notes

ChatGPT-style memory often stores preferences the user typed. Muse relationship profiles go further by structuring contacts into pages on a schedule. That design fits Meta's social graph history, which is why journalists flagged it.

Helpfulness examples in Meta's statement rely on remembering plumbers, invoices, and flower preferences. Those same mechanisms can encode conflict history and closeness scores. Agencies should demo the Strengthening section carefully so clients see social inference risk.

When you write acceptance criteria, prefer evidence-only pages over rich guesswork. Empty fields are safer than invented birthdays. Document that product rule so QA knows what to test.

Sample client talking points on Muse relationship profiles

Use short lines. Clients remember short lines. Here is language India and USA agency delivery teams can adapt without overclaiming.

Muse can build Muse people pages for contacts on an hourly schedule using connected evidence.
Meta says users can wipe memories, disconnect services, and review an audit log.
Researchers say third parties who never signed up may still get profile pages.
We will not connect finance or health tools until wipe and disconnect drills pass.
Relationship strengthening suggestions stay off HR and legal workflows unless counsel approves.

Walk through one harmless contact example, then one sensitive example. The contrast teaches faster than a slide deck. End by assigning an owner for weekly audit-log review.

If a client wants Muse for productivity only, still inventory AI agent relationship data. Productivity connectors are how relationship graphs grow. Your job is to make that path visible before go-live.

Acceptance tests for Meta Muse privacy pilots

Connect one low-risk calendar, then confirm Muse relationship profiles remain sparse for people with no shared evidence.
Ask Muse what it knows about a colleague, then verify claims against connected sources only.
Trigger a forget request and check whether chat transcripts still hold the removed note.
Disconnect the calendar and confirm new facts stop landing on people pages.
Export or screenshot the audit log for the pilot window and attach it to the delivery ticket.

Pass or fail each test in writing. Do not rely on verbal demos alone. India and USA clients both ask for that paper trail during security review.

Key Takeaways

Muse relationship profiles can create a page for every person in a user's life on an hourly process.
Pages may include Facts, History, The relationship, In common, Open threads, and Strengthening.
Instructions say use only available evidence; invented details are worse than an empty page.
Meta Muse privacy claims include a per-user VM, wipe/disconnect controls, confirmation before email or purchase, and an audit log.
Future Society reports pages can cover people who never signed up for Muse.
India and USA agency teams should brief third-party profiling risk before connector-heavy pilots.
Separate consumer Muse experiments from production delivery systems and gadget hardware tracks.

What Debugged Pro recommends this week

Do not block client curiosity. Do block unmanaged connectors on agency-managed devices. Put Muse memory and Muse people pages into your privacy intake form now.

If you already published Muse hardware or small-business posts for the same client, send this privacy brief as a follow-up. One angle per email keeps stakeholders calm. Offer a short workshop that walks through wipe, disconnect, and audit-log proofs.

Need a delivery partner to turn this into SOW language and test scripts? Browse the Debugged Pro blog for related agent privacy posts, then book a consult when you want Muse relationship profiles risk mapped to your India or USA client stack.

About the author: Nitin Kumar is Lead Project Coordinator at Debugged Pro. He works with software and IT agencies across India and the USA on AI delivery, privacy reviews, and client-ready checklists.