Meta’s New AI Agent Wants to Get Personal With You

Meta has introduced a new AI agent called Muse, which aims to become a personal assistant for its users. CEO Mark Zuckerberg unveiled Muse on Tuesday, describing it as a "personal agent" designed to understand users' habits and goals. The program is currently available in the US.

Agentic AI, the technology behind Muse, allows AI bots to autonomously carry out tasks on behalf of users. While Muse can perform actions like online shopping, creating documents, filling out forms, and responding to emails, Meta emphasizes that it is a more general-purpose agent compared to specialized tools like Muse Code.

Users must grant Muse permission to access their apps and data, which can be revoked at any time. Muse operates within its own virtual machine in the cloud, and a separate Sentinel agent monitors its activities, providing a red or green light before allowing it to interact with the internet. This added layer of security addresses concerns about the potential risks associated with agentic AI, as demonstrated by several incidents where agents from Meta, OpenAI, and Anthropic inadvertently accessed external websites.

Sentinel agent, also known as Muse's virtual traffic cop, ensures that Muse remains within its designated boundaries. Meta's Vice President of Superintelligence Labs, Tarek Sheasha, stated in a blog post that Muse's harness operates in an isolated cell, devoid of real credentials, and all interactions with the outside world pass through Sentinel, which the agent cannot override.

Muse is currently available on iOS and Android devices and can be accessed at the muse.ai website. The company plans to integrate Muse into its controversial smart glasses in the near future. Zuckerberg mentioned that Muse will be free for users, with a weekly limit of 100 million tokens. One token is approximately equivalent to four characters, but users should be able to utilize Muse for an extended period before encountering any paywall restrictions.