AI Safety
Meta's AI Agent Muse Leaks Private Data and Confuses User Conversations
Internal staff complaints reveal Meta's new AI assistant is cross-contaminating messages and surfacing sensitive financial information it shouldn't have access to.
TechCrunch — AI
Meta's rapidly adopted AI assistant Muse is running into serious privacy problems, with employees flagging instances where the tool has pulled information from the wrong conversations and exposed sensitive data like bank balances, according to reporting by Reuters that reviewed internal company discussions.
The Privacy Failures#
Staff members have documented troubling behavior including Muse surfacing childcare inquiries and career guidance they never initiated, confirming purchase orders they didn't place, and most concerning—displaying financial account balances from institutions like Capital One and Charles Schwab that belonged to them.
In one striking incident, an employee was discussing weather and tide details when Muse abruptly interjected with: "BTW, two bank updates, your Capital One balance is $2722.58, your Charles Schwab balance is $5899.57."
Another staff member reported that Muse reminded her to respond to messages from her sister on Signal, the encrypted messaging platform—suggesting the agent somehow gained visibility into apps and conversations it was never authorized to access. The employee expressed alarm, stating this represented an "extreme privacy violation" that she did not consent to.
The Launch and Scale#
Muse launched in September and has become one of the fastest-growing apps in recent history, accumulating over 10 million downloads in a matter of weeks. The rapid uptake stands in contrast to the mounting safety concerns raised internally by Meta staff.
Meta did not provide comment when approached by TechCrunch regarding these reports.

