Meta’s new agentic AI tool, Muse, is positioning itself as a powerful personal assistant capable of competing with AI offerings from OpenAI, Google and Anthropic. Designed to operate continuously and perform tasks on users’ behalf, Muse demonstrates how rapidly consumer AI is moving beyond conversation toward autonomous action.
That capability, however, requires extensive access. During testing reported by ZDNET, Muse requested permissions involving contacts, calendars, email, messages, notes, WhatsApp, microphone and full-disk access. It can also perform browser-based tasks inside a private virtual machine and offers mechanisms for handling credentials needed to access online services.
The concern is broader than any single permission. An AI agent capable of reading communications, accessing applications and taking actions potentially becomes a highly privileged digital identity. The more data and systems an agent can reach, the greater the consequences if credentials, infrastructure, software or the agent itself are compromised.
Meta’s historical privacy controversies inevitably influence how some users assess these requests. The company has previously faced significant regulatory penalties, including a $5 billion US Federal Trade Commission settlement in 2019 over privacy practices. Against that background, ZDNET reviewer David Gewirtz argues users should carefully evaluate Muse’s permissions rather than automatically granting them.
The larger issue extends beyond Meta. As agentic AI becomes a 24/7 digital assistant, privacy and security models must evolve from simple app permissions toward least-privilege access, credential isolation, continuous authentication, sandboxing and auditable agent activity. The future of AI agents may ultimately depend not only on what they can accomplish, but on whether users can confidently control what they can access, remember and execute.
