Meta's Muse AI Agent Has a Serious Zero-Day Flaw

Ars Technica reporters discovered a serious zero-day vulnerability in Meta's new Muse AI assistant. A simple ClickFix attack lets an attacker take full control of the agent, and that's just one of several ways to hijack it.

The core problem is that Muse was granted unusually broad privileges from the start — access to personal data, actions, and connected systems. That means hijacking the agent doesn't just expose one feature, it hands over everything it was trusted with, from messages to financial operations.

Check what permissions each AI agent in your company actually has: access should match the role, not be granted 'just in case.' Permissions in 1Cifer follow the company structure — an accountant and a manager see different data and actions, which limits the damage from any single failure.