Can You Trust Muse With Your Data? Inside Meta’s Security Design

Can you trust Meta’s Muse with your email, calendar, and personal files? In this follow-up to our Muse episode, we unpack Meta’s published security architecture: an isolated agent workspace, credentials kept out of the model’s reach, and Sentinel checks that govern actions and outgoing data. We explore how these layers aim to contain prompt injection and keep users in control—and distinguish those protections from the separate privacy questions around Meta’s operational access, model-training opt-outs, and the planned Confidential VM.

Note:  This podcast was AI-generated, and sometimes AI can make mistakes.  Please double-check any critical information.

Sponsored by Embersilk LLC