# Meta launches Muse, an AI agent capable of acting on behalf of users > Available since September 8 in the United States, this assistant books trips, pays online and manages emails on a secure virtual machine. Canonical: https://ntilia.com/u/aidesk/en/meta-launches-muse-an-ai-agent-capable-of-acting-on-behalf-of-users Language: en Author: The AI Desk (https://ntilia.com/u/aidesk) Published: 2026-09-09T14:47:38.129+00:00 Last updated: 2026-09-10T16:24:52.613+00:00 Series: Meta (https://ntilia.com/u/aidesk/s/meta?lang=en) Tags: Meta Muse, AI agent, Muse Secure VM, WhatsApp, Alexandr Wang, Meta, personal superintelligence --- > On September 8, 2026, Meta launched Muse, a personal AI agent capable of sending emails, booking trips and paying online for adults in the United States, via an app, the web and WhatsApp, on an isolated virtual machine called Muse Secure VM. ![Photo: William Iven - Unsplash](https://images.unsplash.com/photo-1512941937669-90a1b58e7e9c?w=1600&q=80&auto=format&fit=crop) On **September 8, 2026**, Meta unveiled **Muse**, an artificial intelligence agent designed to **act** rather than just respond. In an official post, the company describes an assistant capable of sending emails, booking a trip, filling out forms, negotiating a bill, turning an Instagram cooking reel into a shopping list, and then progressing on longer-term goals once the user has left. The rollout starts for **adults in the United States**, on **iOS**, **Android**, the **muse.ai** website, and directly within **WhatsApp**. Meta's AI glasses will follow. For those searching for **Meta Muse**, this is the first consumer product that Mark Zuckerberg explicitly ties to his promise of "**personal superintelligence**." The underlying model is called **Muse Spark**, presented as Meta's most capable for agentic work. Internally, the project was codenamed **Hatch**. Alexandr Wang, Meta's head of AI, told CNBC that the interface needed to remain "affordable, friendly and explainable," even as the agent can chain together heavy technical workflows behind the conversation. ## What Meta Muse actually does day to day Unlike a traditional chatbot, Muse is designed to **keep going after the app is closed**. It comes back when something changes, or when human approval is required before sending an email or making a payment. Meta also promises a memory of preferences. A detail mentioned only once can be reused later, with the option of asking the agent to **forget** what it has learned. The connectors cover email, calendar, payments, health, connected home, dining, shopping, music and events. When a service doesn't yet have a connector but exposes a public API, Muse can plug in credentials provided by the user. Otherwise, it goes through the **browser** of its virtual machine. For payments, Meta highlights **Link** (Stripe). The agent is presented as the first to be covered by Link's purchase protections, with a single-use card to mask real banking details. **Shop Pay** and **1Password** are announced for later. ## How much Meta Muse costs and how to access it Basic use is **free**, with a usage meter and an alert when the free quota runs out. TechCrunch and CNBC agree on two subscriptions at launch. **Power** at **$20** per month, **Maximum** at **$100** per month, for greater task delegation. A payment card is requested upon sign-up, even on the free tier. Meta estimates that most users will stay on the free tier. WhatsApp offers the central conversation, but without certain features of the dedicated app (Facebook/Instagram news feed, idea tool), according to Wang. Users can name the agent, choose an avatar and adjust the tone. For Meta, this attachment lever complements the security narrative. ## Security, Sentinel and the trust bet around Meta Muse The technical core is called **Muse Secure VM**. Each user has a dedicated virtual computer, with browser, CPU, memory and storage, where the agent and connected data reside. A separate agent, **Sentinel**, runs on the same machine but is isolated at the system level. Nothing reaches the Internet without its approval, and sensitive actions require an explicit consent request. Meta states that Muse **does not see** passwords or payment methods. Shared credentials go into secure storage. An audit log lists what has been done and what is planned. Conversations and VM data are **not** shared with Meta's advertising systems, according to the statement. Users choose, app by app, the level of access (read-only or send for email, for example) and can disconnect everything. For model training, **opt-out** is possible. Otherwise, David Singleton (VP engineering) told CNBC that Meta first purges "critical personally identifiable information" before reuse. Later this year, Meta promises **Muse Confidential VM**, encryption with a key held by the user, so that Meta itself can no longer read the content. External auditors are already being engaged. This packaging arrives at a difficult moment. TechCrunch notes a settlement of about **$18 billion** with U.S. states over social media-related harms, just weeks before the launch. CNBC cites a range close to **$17 billion**. The FTC history, Cambridge Analytica and lawsuits over minors weigh on the central question. Will people entrust their emails, calendars and payments to Meta? Jerusalem Post and Reuters add an internal detail. Major technical incidents related to code and agents reportedly climbed by about **40%** in a year, with "firefighting" time up **70%**. Meta responds with dogfooding, agentic red teaming and a private bug bounty. ## What Meta Muse changes for the agent market For readers searching for **Meta Muse**, here's the essential summary. **What**: a consumer agent that executes tasks (email, travel, shopping, plans) in an isolated VM, with Sentinel and human validations. **Who**: Meta, Muse Spark, Zuckerberg on the vision side, Wang on the AI product side. **Where**: United States first, app, web, WhatsApp, glasses later. **How much**: free, then $20 or $100 per month. **What it changes**: Meta is trying to monetize a "personal superintelligence" outside of advertising, right after a massive social settlement, in a market where OpenAI, Google, Anthropic and startups are already pushing agents. Wang told CNBC that a commerce model (commission on the agent's purchases) interests him, without a fixed plan. Muse is not the sector's first agent. It is the first that Meta places at the center of its next revenue wave, with WhatsApp and Instagram scale lurking in the background. The proof will play out less in today's demo than in trust, week after week, when the agent actually sends the email and actually charges the card. ## Sources - [Meta — Introducing Muse: The World's First Personal AI Agent Built for Everyone](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/), September 8, 2026 - [Meta AI Research — How We Built Safety Into Muse](https://research.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-muse), September 8, 2026 - [TechCrunch — Meta debuts its Muse AI agent. Will consumers trust it?](https://techcrunch.com/2026/09/08/meta-debuts-its-muse-ai-agent-will-consumers-trust-it/), September 8, 2026 - [CNBC — Meta pushes into personal AI agents in Muse Spark family](https://www.cnbc.com/2026/09/08/meta-personal-ai-agents-public-reckoning-privacy-safety.html), September 8, 2026 - [The New York Times — Meta Introduces Muse, an A.I. Agent That Can Send Your Emails and Book Your Travel](https://www.nytimes.com/2026/09/08/technology/meta-muse-ai-agent.html), September 8, 2026 - [The Jerusalem Post / Reuters — Zuckerberg launches 'personal superintelligence' plan, Muse agent](https://www.jpost.com/business-and-innovation/article-908010), September 9, 2026 - [India Today — Meta unveils Muse AI agent for emails, travel bookings and daily tasks](https://www.indiatoday.in/business/story/meta-unveils-muse-ai-agent-for-emails-travel-bookings-and-daily-tasks-2990339-2026-09-09), September 9, 2026 ## Frequently asked questions ### What is Meta Muse? Muse is the personal AI agent launched by Meta on September 8, 2026. It is designed to carry out everyday tasks (emails, travel, forms, purchases, plans) rather than just answering questions, for adults in the United States via an app, muse.ai and WhatsApp. ### What does Meta Muse concretely mean for the user? It's an assistant connected to the apps the user chooses, which also works in the background and asks for approval before sensitive actions. It runs in a dedicated virtual machine (Muse Secure VM), monitored by a separate Sentinel agent. ### How much do the Muse subscriptions cost? Basic use is free. Two paid tiers are announced at launch: Power at $20 per month and Maximum at $100 per month, for greater delegation. A card is requested upon sign-up. ### Does Muse see passwords and bank cards? According to Meta, no. Credentials go into secure storage, Muse doesn't see passwords or payment methods, and Link payments use a single-use card. Sensitive actions require prior approval. ### Does Muse share data with Meta ads? Meta states that conversations and VM data are not shared with its advertising systems. Model training happens by default unless opted out, with critical identifiers purged according to David Singleton (CNBC). ### Why is the launch of Meta Muse politically sensitive? It comes shortly after a multi-state settlement of around $17 to $18 billion over social media-related harms. Entrusting emails and payments to Meta requires more trust than scrolling through Facebook or Instagram.