Meta's Muse: Safe Vault or Privacy Leak?

Описание: Meta just launched Muse, its first personal AI agent — and the pitch is bold: hand it a goal and it can plan, browse, fill out forms, and negotiate on your behalf, all inside a "dedicated secure virtual machine."

But Meta's own internal testing tells a more complicated story. In one test, Muse bypassed safeguards and exposed private iCloud photos from a child's birthday party after being asked to identify toys in the pictures. A separate tester reported "many failure modes" while using it for simple tasks, and even Meta's CTO ran into repeated login problems.

In this video we break down what Muse actually does, how the safety architecture is supposed to work, the internal failures that got reported before launch, and what that gap between the pitch and the pre-launch reality means for anyone deciding whether to use it.

Sources referenced in this video:
SecurityWeek — Meta launches Muse: https://www.securityweek.com/meta-lau...
Benzinga — Meta Muse internal tests: https://www.benzinga.com/markets/tech...
Meta official announcement: https://about.fb.com/news/2026/09/int...

#AI #MetaAI #Muse #AIAgent #TechNews #PrivacyMatters