Meta Releases Muse, a Personal AI Agent With Privacy ‘Built Into It’

Meta has launched Muse, a personal AI agent designed to help users automate everyday tasks on the internet and within Meta's products. The AI assistant is now available for iOS and Android users through a dedicated app, as well as on the Muse.ai website. Users can also interact with Muse directly through WhatsApp. In the future, Meta plans to integrate Muse with its AI glasses.

Muse is free to try, but users interested in automating numerous digital tasks will require one of Meta's subscription plans. The AI agent is the brainchild of Meta Superintelligence Labs, an AI unit formed by CEO Mark Zuckerberg's team to compete with OpenAI, Anthropic, and other players in the AI field. The unit has been testing Muse internally under the codename "Hatch," and employees have been using it to perform tasks like operating third-party applications and browsing the web on their behalf.

Users can interact with Muse using natural language, and it can perform tasks like sending emails, booking travel, and even helping sell a car. Meta has also integrated a secure payment system called Stripe Link into Muse, which generates a single-use card number to protect users' financial data. The agent is built on an architecture called Secure VM, designed to isolate each user's activity in a virtual machine to prevent untrusted data from the web and integrations from compromising the agent's ability to act on behalf of the user.

Meta has taken significant steps to ensure user trust and data security with Muse. The Secure VM architecture includes a Sentinel that monitors data movement and prompts users for approval when actions occur outside the virtual machine. Users can also opt out of having their data used for training Muse. Meta plans to offer a more private version called Confidential VM, where users manage their own access keys, and Meta will not have access to the user's agent VM.

Meta is emphasizing the privacy and security features of Muse to differentiate itself from competitors. The company has already conducted extensive testing of the Secure VM architecture, including by its human and agentic red teams and through bug bounty programs. Users can expect Meta to publish the Confidential VM binaries and a transparency log to further assure them of the product's privacy guarantees.