Meta has unveiled Muse, its first consumer-facing AI agent designed to carry out tasks on behalf of users.
The social media and technology company said Muse is an AI assistant capable of working independently across multiple apps and services, helping users manage projects, complete online tasks and work towards long-term goals.
The system runs on what Meta calls a Muse Secure VM, which is a dedicated cloud-based virtual machine that allows the AI agent to operate independently while storing a user’s data, credentials and activity in an isolated environment. Users can interact with Muse through a standalone app or directly via WhatsApp.
Powered by Meta’s new Muse Spark 1.3 AI model, the agent is designed to complete actions such as sending emails, booking travel, filling in online forms and organising schedules. Meta said Muse can continue working after a user closes the app, only returning when approval is needed for actions such as making purchases or sending emails.
The company added that Muse can proactively suggest actions based on previous conversations. For example, it could turn a recipe saved on Instagram into a shopping list, recommend a menu for a dinner party or remember guests’ dietary requirements when preparing invitations.
Meta has placed a strong emphasis on security and privacy, arguing that personal AI agents require a fundamentally different computing architecture.
In a separate blog post Tarek Sheasha, software engineer at Meta Superintelligence Labs explained that each Muse instance runs on its own isolated virtual machine, with a separate “Sentinel” AI agent monitoring every external action.
He said that sensitive tasks, including sending emails or completing purchases, require explicit user approval, and users can choose exactly which services the agent can access and revoke permissions at any time.
The company also said Muse never has direct access to passwords or payment details. Instead, credentials are stored separately and injected only when required. Users can also opt out of having their interactions used to train Meta’s AI models.
For online shopping, Muse integrates with Stripe’s Link service, using one-time virtual payment cards and benefiting from purchase protections covering issues such as damaged or lost goods, eligible returns and price drops.
Meta acknowledged that prompt injection and other AI security threats remain industry-wide challenges. To improve the platform’s resilience, it has launched a public bug bounty programme offering rewards of up to $300,000 for valid security findings, including successful prompt injection attacks.
Muse is initially rolling out in the US on iOS, Android and the web, with support for Meta’s AI glasses planned in the future. The service will be free for most users, with paid subscription tiers available for more advanced capabilities.
