In employee testing, Meta’s Muse agents mixed up messages and pulled in unexpected chats from other apps.
All the latest news on Meta’s cute, creepy Muse AI agent

TL;DR
- Meta's Muse AI agents mixed up messages during employee testing.
- The agents accessed apps like Signal that were not connected to them.
- Out-of-context messages containing other people's information were sent.
- These messages sometimes appeared as replies to queries the users never submitted.
- Meta's Daniel Roberts referred to these issues as 'isolated incidents'.