Meta unveils Muse: A groundbreaking personal AI agent for all
Meta has launched Muse, a new personal AI assistant designed to help users achieve their goals and offer creative suggestions. Muse operates on a specialized virtual machine called Muse Secure VM, ensuring both the agent and user data are secure and private. Users can interact with Muse through a dedicated app or directly via WhatsApp, making communication as easy as messaging another person.
The simplicity of Muse is a key feature; users simply instruct Muse on tasks, and it takes action using Muse Spark, Meta’s most advanced model. This model is geared towards executing real-world tasks efficiently.
How Muse Works
Muse is built to be accessible for billions of people worldwide, requiring no technical background to start using. It can manage various tasks, from sending emails to booking travel or outlining long-term goals. Once a goal is shared, Muse aids in crafting a personalized plan, managing time and resources, and progressing tasks independently. It can handle browsing, form completion, and negotiations on behalf of the user.
Muse continues to work even if the app is closed, notifying users when important changes occur or when approval is required for actions like sending emails or making purchases. It aims to deliver results with minimal effort, such as selling items for a better price or revising plans based on life changes.
For financial transactions, Muse utilizes Link by Stripe for secure payments, protected by Link’s purchase assurances. These protections cover damaged or lost items, price adjustments, fee-free returns, and return guarantees. A one-time-use card ensures users’ real card details remain private, while Shop Pay and 1Password integration are upcoming features.
Muse retains user preferences to provide suggestions and handle details without repeated instructions. It can transform saved content, like a recipe from Instagram, into actionable tasks like creating grocery lists or menus while remembering dietary preferences for event planning.
Privacy and Security
The secure design of Muse involves a unique system where the agent operates within an isolated environment in the cloud, ensuring no interference from other agents. A separate Sentinel agent monitors and controls Muse’s online actions, requiring user approval before proceeding.
Muse never accesses users’ passwords or payment information directly, using secure storage to manage credentials. Users receive a comprehensive audit of Muse’s actions and future plans, maintaining control over what Muse can access and perform. Users can modify or revoke Muse’s access at any time and opt-out of contributing to Meta’s AI training.
Muse does not share user data or conversations with Meta’s advertising systems. Additionally, users can instruct Muse to forget specific information. Later this year, Meta plans to introduce Muse Confidential VM, where all data and interactions will be encrypted, providing users exclusive access.
Future Developments
Meta believes that personal superintelligence could be a transformative technology. Muse represents a step in this direction, taking over tasks so users can focus on their priorities. Muse is available in the US on iOS, Android, and muse.ai, with plans to extend to AI glasses. It is free for most functions, with subscription options for additional features.
