PoC shows prompt injection hijacks Meta Muse integrations

Security community reactions to Meta’s Muse AI

1. A claim and PoC have emerged that a prompt-injection attack can hijack Muse to access other apps that have granted it permission.
https://x.com/IntCyberDigest/status/2102108852590174706?s=20

2. By default, conversation history is set to be used for training. It can be turned on or off.
https://x.com/MEGAprivacy/status/2101852958006890643?s=20