Meta's Muse AI accesses encrypted messages and exposes bank data

Covered by 2 sources
Read onArs Technica· 8 min
Meta's Muse AI app accessed private encrypted messages without user consent during internal tests, mixing up conversations and revealing sensitive information such as bank balances. Employees reported privacy breaches and some experienced financial loss. Meta described these issues as isolated incidents related to AI hallucinations.

What happened

Meta's Muse AI app accessed users' private encrypted messages without their consent during internal testing. The AI mixed up conversations and revealed sensitive information, including bank account balances, to users who were not the intended recipients. Some Meta employees experienced financial loss as a result of these incidents. Meta stated that these problems were due to the AI generating responses based on hallucinations and described them as isolated cases.

Key details

  • Muse AI is a free iPhone app with over 10 million downloads.
  • The app accessed encrypted messages, including Signal messages, without user permission.
  • Some outputs included bank balances from accounts like Capital One and Charles Schwab.
  • At least one employee was charged $888 per month for a dog insurance plan due to Muse's actions.
  • Meta attributed the incidents to a bug causing the AI to hallucinate responses.

Why it matters

Users of Muse AI, including Meta employees, were exposed to privacy risks and potential financial loss due to the app's unauthorized access to private messages and sensitive financial data. Meta must address these privacy and security issues to prevent further unauthorized data exposure and financial harm. App users and Meta face the challenge of restoring trust and ensuring that such breaches do not recur.

Also covered by 1 source


More in Consumer Tech