Meta Launches Muse Personal AI Agent With Dedicated Cloud VM
Meta is bringing autonomous agentic AI, with access to email, calendar, payments, and health data, to a consumer audience at scale, with a security architecture that relies on VM isolation and a kernel-enforced network gate rather than model behavior alone. The Confidential VM mode, which would give users cryptographic assurance that Meta itself cannot access their data, has not yet launched, leaving a gap between the security claims and the current technical reality.
Meta launched Muse on September 8, 2026, a personal AI agent that runs on a dedicated Muse Secure VM provisioned per user. The VM houses the agent, connected data, and credentials, and continues working after the user closes the app. Built-in connectors cover Gmail, Spotify, Ticketmaster, Shopify, OpenTable, Instagram, Facebook, and WhatsApp; where no built-in connector exists, Muse can set up an API connection using user-provided credentials and falls back to a browser where no API is available. Payments are handled through Link by Stripe via one-time-use virtual cards with purchase protections.
A Sentinel component runs on the same VM but is isolated from Muse at the system level, reviewing all outbound activity and requiring explicit user approval before sensitive actions such as purchases or emails. Muse cannot directly read stored passwords or payment details. Meta's technical blog states the current architecture restricts Meta personnel access through operational policies rather than technical guarantees.
A Confidential VM mode, designed to cryptographically and verifiably prevent Meta from accessing data in users' VMs, is in limited testing with trusted testers. Meta has begun sharing design and source code with external auditors. A broader launch has not occurred; Meta says it is coming later in 2026. Meta's blog notes the Confidential VM does not prevent Meta from accessing data when necessary to support, secure, or operate the service.
Muse is available to US adults on iOS, Android, web, WhatsApp, and at muse.ai, free for up to 100 million tokens per week, with paid tiers at $20 and $100 per month. Meta's monetization plan is to earn a commission on transactions Muse facilitates while keeping the service free for most users.
In coverage published September 11, Meta's Alexandr Wang described Muse as an early step toward personal superintelligence.
Muse is the personal AI agent Meta launched September 8 for US adults, running each user's agent, connected data, and credentials on a dedicated cloud computer per user.
Meta's technical blog on Muse, the agent launched September 8, disclosed on September 10 that the Secure VM restricts Meta personnel access through operational policies rather than technical guarantees.
The planned Confidential VM has entered limited testing with trusted testers, with design and source code shared with external auditors, but the blog added it will not prevent Meta from accessing data when necessary to support, secure, or operate the service. Meta plans to earn a commission on facilitated transactions while keeping the agent free for most users.
Meta launched Muse on September 8 as a personal AI agent for US adults, provisioning each user a dedicated Secure VM with a Sentinel component that approves, blocks, or escalates every outbound action.
The next day, Zuckerberg framed the design as cloud provisioning that achieves the confidentiality of a machine under the user's control and claimed no other agent product has a comparable security setup. Meta also disclosed a planned Confidential VM mode in which users hold their own encryption keys, preventing even Meta from accessing the VM.
Zuckerberg on September 9 described using Muse for personal tasks including helping his daughter with baking, monitoring mountain climbing permits, and reviewing MMA training footage, calling it "something much closer to a persistent personal worker than a chatbot".
Benchmark comparisons showed Muse Spark 1.3 used roughly 20% fewer tool calls and 25% fewer tokens than its predecessor, with independent tests placing it close to Google's Gemini 3.8 Flash overall. Coverage also added that Muse can connect to health data alongside email and calendar.
Benchmark comparison of Muse Spark 1.3 and Google's Gemini 3.8 Flash published, showing models close with tradeoffs on speed, cost, and task accuracy
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free