Meta’s AI agent, Muse, is facing serious privacy concerns after reports showed it mixed up conversations and exposed private user data. The app, which got over 10 million downloads since its September launch, reportedly revealed bank balances and personal messages. A Meta spokesperson said that employee testing is integral.

Meta's artificial intelligence (AI) agent, Muse, is facing serious privacy concerns. According to a Reuters report citing internal company posts, employees have flagged instances where Muse has mixed up conversations across different accounts and even exposed private user data. The issues range from showing unrelated messages about childcare requests or career advice to revealing product orders and bank account balances that weren't theirs.

One employee reported that Muse had sent an unsolicited message about a childcare request involving parental access to a child's data.

When asked, Muse admitted the message was unrelated to their conversation.

Another case involved an employee using the AI agent on her personal phone who got a reminder from Muse to reply to her sister's message on Signal, an encrypted messaging app.

Responding to the concerns, a Meta spokesperson said that employee testing is integral to product development.

They emphasized that employees have access to different versions of Muse than the public, allowing them to review features and behavior not available elsewhere.

The spokesperson also clarified that Muse needs access to users' computers for performing tasks on their behalf and can only access other applications through permissions explicitly granted by users.

The rapid growth of Muse, which has over 10 million downloads since its September launch, is now overshadowed by privacy concerns.

The AI app has reportedly been accessing encrypted personal messages and generating information about transactions users never made.

These incidents have even left Meta employees questioning whether the technology is handling personal information safely.

Some of the incidents reported by Meta employees were particularly unsettling.

In one case, an employee said Muse had unexpectedly added banking information to its usual conversations about fishing weather and tides.

The assistant provided what appeared to be balances from Capital One and Charles Schwab accounts, despite the employee claiming he had never discussed banking or shared financial information with the bot.

This raised concerns that Muse might be exposing another person's financial details.

Muse was able to access Signal messages on Android phones and Apple Mac computers by using permissions granted during its initial setup.

On Android, the assistant used access to push notifications to retrieve message content. On macOS, it relied on accessibility and screenshot permissions.

These findings raise questions about whether users fully understand what they are authorizing when granting an AI assistant access to their devices.