AI
Report says Meta’s internal Muse tests had account mix-ups and private data exposure
After reviewing internal Meta posts, Reuters says employees were concerned that the AI agent Muse was mixing up conversations from different accounts and exposing private data. Anomalies included parenting and career advice and orders the user had not requested, as well as bank balances that did not belong to the user; one employee also said reminders to reply to Signal messages were sent to a personal phone. Meta responded that the employee-tested version differs from the public version and is intended to gather feedback and improve safety and privacy.
TechCrunch, citing Reuters, reported that some Meta employees described in internal posts how Muse would surface parenting requests they had never made, career advice they had never sought, product orders they had never placed, and bank-account balances that did not belong to them. One employee said Muse suddenly inserted Capital One and Charles Schwab account balances into a chat about weather and tide information.
Another employee, while using Muse on her personal phone, received a reminder saying she should reply to a message from her sister on the end-to-end encrypted messaging app Signal. She said in an internal post that she had not granted Muse permission to access other apps and did not remember being asked for such authorization, calling it a serious privacy violation.
A Meta spokesperson told TechCrunch that employee testing is a core part of product development and is precisely intended to obtain feedback so that safety and privacy protections can be implemented and the product improved. He emphasized that the Muse version available to employees differs from the public version and often includes features and behaviors not yet offered externally. Muse launched in September, and the report says it has surpassed 10000000 downloads, making it one of the fastest-growing apps.