Meta Muse profiles everyone you know, customers included

Meta's Muse AI agent keeps a profile on every person in your life, and it updates them every hour. WIRED reported on October 4 that system prompts pulled out of Muse tell it to maintain "a page for every person in the user's life." If anyone on your team has connected Muse to work email or messages, that list includes your customers, your vendors, and the people you're negotiating with.

What actually happened

A researcher got Muse to share its own instruction files through the normal chat interface and gave them to WIRED. According to WIRED's reporting, summarized by Social Media Today and Techmeme, the instructions tell Muse to:

• Keep pages on family, partners, friends, colleagues, "collaborators," and people the user follows
• Record where they live, what they do, and "the threads that recur"
• Track "dates that matter," such as birthdays and anniversaries
• Log history: a trip, an argument that got resolved, a recent milestone
• Refresh all of it on an hourly cadence and suggest ways to strengthen each relationship

The people being profiled never agreed to any of this, and there's no setting that lets them opt out. That's the part privacy advocates are focused on.

None of this is a bug. It's how the product is designed to work. A personal agent that remembers who you know is useful. It's also a contact database that you didn't build, can't fully inspect, and that's run by a company that makes almost all its money from advertising.

Why Meta Muse profiles matter for your business

Meta is actively selling Muse to small businesses, and staff connect personal agents to work accounts because it saves them twenty minutes a day. Do the math on what follows. Every customer thread, supplier dispute, and quote negotiation in that inbox becomes material for a profile page on someone who never dealt with Meta. You may have promised those people confidentiality in a contract or a privacy policy. Your employee's agent didn't sign it.

This is a data-governance problem, not a Meta problem. Any agent with long-term memory and inbox access will build something like this. Muse just had its instructions exposed.

What we'd do this week:

1. Inventory agent connections. Ask the team which AI agents have access to work email, Slack, or the CRM. Put the answers in a sheet.
2. Separate personal from work. Personal agents don't get work accounts. If an agent is going to touch customer data, the business picks it, owns it, and can turn it off.
3. Check your privacy policy and NDAs. If they promise customer data stays with named processors, an unlisted agent breaks that promise.
4. Pick agents whose memory you can read and delete. If you can't export it, you don't control it.

Sources: WIRED, Social Media Today, Techmeme.