Extracted system prompts show Meta's Muse compiles “a page for every person in the user's life”, with facts, history, tips to improve relationships, and more
Millions have downloaded Meta's AI agent Muse. But getting it to do your bidding comes with privacy costs.
Context & Ripple Effects
Muse arrived as an agentic extension of Meta AI, with Gmail and Google Calendar connections intended to turn personal data into task execution. Meta then paired broad free access with paid compute tiers, helping it reach the top of Apple’s US free-app chart within weeks of launch.
Reviews had already identified trust in Meta with personal data as Muse’s central adoption hurdle. The extracted instructions make that concern more concrete: the agent’s usefulness is tied not only to a user’s own information, but to its modeled understanding of people around them.
First-order effects
- Muse users receive relationship-oriented assistance shaped by stored facts and histories about people in their lives, making personal context a core part of the product rather than an incidental input.
- Friends and family members can become subjects of Muse’s analysis without being the account holder; public reaction focused on the absence of consent for people an agent profiles.
Second-order effects
- Meta must persuade users that the convenience of a highly contextual assistant outweighs the privacy implications that earlier Muse reviews had already flagged, particularly as the app’s broad free tier drives adoption.
- ChatGPT and other personal-agent providers gain a clearer privacy-positioning opportunity: relationship memory becomes a product differentiator, but also a liability when it extends to non-users.
Third-order effects
- Personal AI agents are moving from answering prompts to maintaining ambient social models, raising pressure for product controls that distinguish a user’s consent from consent by the people described in their data.
- If these agents normalize inferred profiles of third parties, privacy governance will increasingly center on who may be analyzed by an assistant, not only on what its account holder chooses to share.
The trend: Consumer AI assistants are evolving into ambient memory systems whose value comes from persistent, relationship-level context—and whose privacy stakes extend beyond the user.