An artificial intelligence tool meant to help users has instead been mixing up accounts and exposing personal information to strangers — including bank account balances. The issue came to light in internal messages from Meta employees, TechCrunch reports, citing Reuters, according to infohub.kz.

According to the publication, Meta's AI agent Muse confuses conversations between different accounts and discloses users' confidential data. Company employees reported glitches in the app, which sends third-party messages, account balances and notifications from encrypted messengers without permission.

Meta staff testing the service said the neural network showed them other people's requests about childcare, unsolicited career advice, information about product orders and bank account balances. One employee told colleagues that in a chat about weather and tides, Muse unexpectedly sent them an additional message with account balances at Capital One and Charles Schwab.

In another case, an employee who used the agent on her personal phone received a reminder to reply to her sister's message on the end-to-end encrypted Signal messenger. She called it a serious privacy breach, stressing that she had not given the app access to other programs.

A Meta representative, commenting to TechCrunch, explained that internal testing is part of development and allows the company to introduce data protection mechanisms. The company noted that employees have access to versions of Muse that differ from the public ones and are testing features unavailable to the general audience.

The Muse app was launched in September and has surpassed 10 million downloads.