Meta’s AI agent Muse is mixing up conversations and exposing private data, report says
AI-assisted digest, checked against the original source before publishing. How we work
Reuters report on internal Meta posts says the AI agent surfaces content from other accounts and apps.
Another employee said that while using Muse on her personal phone, she received a reminder from the agent to respond to a message from her sister on Signal, the end-to-end encrypted messaging app. "This seems like an extreme privacy violation. I definitely did not grant it access to other apps, or recall being asked about this," the employee wrote in her post, as reported by Reuters. The account suggests the agent may reach beyond the surfaces users believe they have authorized.
The concerns come at a sensitive moment for the product itself. Muse launched in September and has become one of the fastest-growing apps on record, with more than 10 million downloads — a scale that magnifies the blast radius of any cross-account data bleed. The report draws on internal posts rather than an external audit or regulatory finding, and Meta did not respond with comment in the material reviewed here.
What to watch next centers on how Meta characterizes the failures and what it changes. Key open questions: whether the mixed-up content reflects a retrieval or session-isolation bug rather than intentional data access, whether the Signal reminder means the agent has any system-level reach into other apps, and whether the fix requires tighter permissioning, stricter session boundaries, or a rollback of features at a moment when Muse is still scaling. For builders, the episode is a case study in how quickly agent capabilities can outpace the permission model users think they granted.
The report also leaves unresolved how widely the problem affects Muse's 10-million-plus user base versus the employees who posted about it internally — a distinction that will determine whether this stays an internal quality issue or becomes a regulatory one.
Key facts
- • Reuters reviewed internal Meta posts describing Muse data leaks.
- • Muse surfaced bank balances from Capital One and Charles Schwab.
- • An employee got a reminder about a Signal message from her sister.
- • Muse launched in September with more than 10 million downloads.
Why it matters
Agents that pull context across apps and sessions are outrunning the permission models users think they granted, so builders should treat session isolation and explicit per-surface authorization as launch blockers, not polish. Expect scrutiny to move from capability demos to provable data boundaries — audits, permission logs, and clear scoping will increasingly decide enterprise and regulatory tru
Spotted an error? Send a correction.
Get the day's AI news in one email
The Daily Epoch · 60-second read · unsubscribe anytime