Meta Introduces Muse: A Personal AI Assistant Designed With Privacy Features Integrated

Meta announced on Tuesday the launch of Muse, a personal AI agent designed for messaging to streamline digital tasks within a secure cloud framework, emphasizing security and privacy features that the company claims are “integrated from the start.”

Meta states that Muse is available starting today for iOS and Android users through a dedicated Muse app, as well as on the website Muse.ai. Users can also interact directly with Muse via WhatsApp. Additionally, those using Meta’s AI glasses will soon be able to engage with the Muse agent. While Muse is free to try, users looking to automate numerous digital tasks will need to subscribe to one of the company’s AI plans.

Muse represents Meta’s latest effort to compete with popular AI agents such as OpenClaw and Instinct, which allow users to message them for task automation. Developed by Meta Superintelligence Labs, the AI division established by CEO Mark Zuckerberg about a year ago to rival OpenAI and Anthropic, Muse has attracted researchers with enticing compensation packages. This initiative is based on the premise that AI agents will revolutionize how average users navigate both the internet and Meta’s products.

Previously reported by WIRED, Meta has been internally testing Muse under the codename “Hatch,” enabling employees to use it for autonomous operation of third-party applications and web browsing on their behalf.

According to a blog post from Meta, Muse is designed for immediate usability, claiming “no learning curve.” Users can interact with Muse using natural language, allowing the agent to autonomously handle tasks such as sending emails, booking travel, or even assisting in car sales.

Muse can also make purchases on behalf of users, utilizing a specialized payment structure created by Stripe. This tool, known as Link, generates a single-use card number to prevent agents from entering real financial information across various platforms. Meta points out that Muse is the first AI agent protected by Link’s purchase safeguards, ensuring no-fee returns.

While Meta may be late to the personal agent market, it appears to be aiming for differentiation through a strong emphasis on security and privacy features for Muse. The agent is launching with an architecture termed Secure VM, intended to segregate each user’s activity in a virtual machine, shielding untrusted web data and integrations from the operational part of the agent.

Developing a personal AI agent requires substantial user trust, an area where Meta has faced notable challenges over the years. To encourage users to engage with Muse—and to permit integration with their third-party applications and services—Meta is striving to build confidence that it can manage user data responsibly.

“We recognize that it’s essential to be responsible with personal data when developing a product that accesses various sources of information, so we’ve intentionally designed this system,” remarks David Singleton, vice president of engineering for consumer products at Meta Superintelligence Labs. “Furthermore, we’ve implemented what we refer to as the Sentinel, which monitors data leaving the VM, ensuring it aligns with existing policies granted by the user or system, or prompts a human-in-the-loop dialogue for action approval.”