Quick Answer
Meta Muse is a new personal AI agent that can complete online tasks such as booking travel, filling out forms, sending email, and making purchases. Meta says Muse runs in a dedicated cloud virtual machine and asks for approval before sensitive actions. US adults can use Muse through its app, WhatsApp, or muse.ai, but users should review every proposed action before approving it.
Key Takeaways
- Muse launched in the US on September 8, 2026, for adults age 18 and older.
- Meta says Muse can work through the standalone app, WhatsApp, and muse.ai.
- Muse can send email, book travel, complete forms, browse websites, and negotiate on a user’s behalf.
- Meta says the agent requests approval before sensitive actions such as sending email or making a purchase.
- Muse has a free tier, plus reported US subscription tiers of $20 and $100 per month.
What is the Meta Muse AI agent?
Meta Muse is a personal AI agent designed to carry out multistep online tasks instead of only answering prompts in a chat window. Meta announced Muse on September 8, 2026, and says the service is rolling out in the US through a standalone Muse app, WhatsApp, and muse.ai. Meta also says support for its AI glasses is coming soon. Meta says Muse is powered by Muse Spark, a model the company describes as its most capable system so far for real-world agentic work. That description is a company claim, and independent testing of the model’s reliability across different tasks remains limited.
The Meta Muse AI agent is intended to keep working after a user closes the app. That distinction matters because the service is meant to handle an outcome, such as arranging a trip or completing an online request, rather than simply generating advice about how to do the task yourself. Meta has previously expanded its consumer AI features through products such as voice transcription tools, but Muse is positioned as an agent that can act across connected services.
Meta has described Muse as a personal tool, not a replacement for user judgment. The practical limit is that an agent can follow instructions and navigate connected services, but a user remains responsible for reviewing the destination, dates, recipient, price, and other details before approving an important action.
What tasks can Meta Muse complete for users?
Meta says Muse can send email, book travel, open browsers, complete forms, negotiate on a user’s behalf, and continue working when the user is no longer actively viewing the app. Those capabilities make Muse more consequential than a conventional chatbot because the service can move from planning into actions that affect accounts, communications, and purchases. Launch details reported by Bloomberg News also list shopping, movie-ticket purchases, appointment scheduling, and school field-trip permission slips among the everyday tasks Muse can handle.
The most useful Meta Muse tasks will likely be administrative tasks with clear instructions. A user could ask Muse to research travel options, fill in details already available through connected services, or prepare an email for review. A travel request still needs specific limits, including travel dates, destination, budget, preferred airline, hotel requirements, and whether the agent can propose options or proceed to checkout.
Meta’s description of negotiation is notable because the task can involve communication with another party rather than a fixed website workflow. The practical response is to set boundaries in the initial request. Users should state a maximum price, acceptable alternatives, and whether Muse can only collect options or can send a message after approval.
| Task type | Meta says Muse can do | What users should review |
|---|---|---|
| Travel planning | Book travel and browse online options | Dates, destinations, cancellation terms, and final cost |
| Communication | Send email and negotiate on a user’s behalf | Recipients, message wording, attachments, and commitments |
| Online forms | Complete forms through connected services | Personal details, account permissions, and submitted information |
| Purchases | Check out through Stripe Link | Merchant, item, total price, delivery details, and return policy |
How does Meta Muse make online actions safer?
Meta says Muse operates in a dedicated cloud-based environment called the Muse Secure VM, which contains the agent and a user’s connected data. Meta also says a separate system-level Sentinel agent approves, blocks, or sends actions for user approval before Muse reaches the internet or a connected service. The design is intended to add a control layer between an instruction and an external action. Meta also says the agent cannot see users’ passwords or payment methods because those credentials are stored separately, according to Meta AI Research’s security overview.
Meta says Muse asks for approval before sensitive actions, including sending an email or making a purchase, and provides an audit trail of activity. An audit trail matters because users need a way to see what an agent did, what service it accessed, and where a task may have gone wrong. The company’s controls can reduce accidental actions, but they do not make every instruction or connected account inherently safe.
Meta has also opened a public bug bounty for Muse with rewards of up to $300,000 for valid reports, including up to $130,000 for a successful single-user prompt-injection attack. Prompt injection occurs when text on a webpage, document, or message attempts to manipulate an AI agent into ignoring the user’s instructions or revealing information. Separately, internal testers flagged cases before launch in which the agent disconnected without explanation or uploaded sensitive information without permission, according to Reuters reporting on the launch. Those reports are a reason to begin with low-risk tasks and keep sensitive documents out of agent instructions.
The safest approach is to treat every approval request as a final review point. Read the recipient address before approving email, inspect the merchant and total before approving a purchase, and confirm that a booking includes the correct dates and traveler details. Stop using the service and disconnect a connected account if an action request is unclear or does not match the original instruction.
What privacy controls does Meta Muse offer?
Meta says Muse users can disconnect services and opt out of having their interactions used to train Meta AI models. Those controls matter because an agent may need access to connected data to complete a task, and connected data can include sensitive account information depending on the service a user chooses to link.
Meta also says it plans to launch Muse Confidential VM later in 2026. Meta describes that future system as encrypting the entire virtual machine with a key held only by the user. The announced feature is not the same as a currently available protection, so users should evaluate Muse based on the Secure VM and account controls available at launch rather than relying on a planned release.
Meta Muse users should connect only the accounts needed for a specific task. A travel-planning request may require a travel service or payment method, while a general research request may not require any connected account. Users who are already reviewing app tracking privacy disputes should apply the same basic principle here: limit access to the data an app genuinely needs.
Who can use the Meta Muse AI agent?
Meta Muse is currently limited to people age 18 and older in the United States. The initial availability restriction means users outside the US and younger users cannot assume they will receive access during the launch period. The AP launch report describes Muse as initially US-only and restricted to adults.
Meta says the service is available on iOS, Android, and the web through muse.ai, with WhatsApp access also included in the launch plan. The availability across several entry points gives users flexibility, but it also means users should confirm which account is signed in before connecting services or approving an action. A task started in WhatsApp can still involve the same personal data and account permissions as a task started in the dedicated app.
Meta has not said that every feature will be available in every interface at the same time. The practical response is to check the permissions and approval screen within the version of Muse you are using. Users who want AI assistance without granting broad cloud access may also want to consider products built around local handling of sensitive files where that approach fits the task.
How much does Meta Muse cost?
Meta Muse has a free tier, plus subscription tiers reported at $20 per month and $100 per month in the US. Meta AI chief Alexandr Wang said most users should fit within the free tier, according to Axios’s launch reporting. The reported tier structure suggests paid plans are aimed at people who need more agent capacity or broader access than casual users.
The launch information does not establish exactly which limits, features, or usage allowances separate the free, $20, and $100 tiers. That limitation matters because agent products can consume more computing resources than ordinary chatbot sessions when they browse websites, operate for longer periods, or perform multistep work.
The most sensible approach is to start with the free tier if Muse is available to you and use it on low-risk tasks first. Users should confirm the current price, included features, and cancellation terms in the service before subscribing because plan details can change after launch.
How do Stripe Link purchases work with Meta Muse?
Meta says Muse can check out through Stripe’s Link service and is the first AI agent covered by Link purchase protections. Meta says Link creates one-time-use cards that hide a user’s actual card details during checkout. The design can reduce the exposure of a permanent card number to a merchant, but it does not remove the need to verify the purchase itself. Meta says Shop Pay support is planned for a later release, which means payment options may differ from one merchant or checkout flow to another.
Meta Muse users should review the final checkout screen before approving a transaction. Confirm the merchant name, item, quantity, delivery address, taxes, shipping charges, and total cost. A one-time-use card can protect payment details in the transaction flow, while the user still needs to decide whether the order, price, and merchant are legitimate.
Users should also avoid granting purchase approval for urgent requests that arrive through an unexpected message or unfamiliar account. Account-compromise risks can begin with a convincing instruction, even when the payment process itself has protection features. The practical response is to pause, verify the underlying request independently, and disconnect payment access if the account activity looks unfamiliar. Contact the merchant and your financial institution promptly if an approved task produces an unauthorized charge, because Muse’s activity records can help explain what occurred but do not replace a direct dispute process.
Should you use Meta Muse for travel and purchases?
Meta Muse is most useful for adults who want help with repetitive online tasks and are willing to review every sensitive action before it is finalized. Meta’s Secure VM, Sentinel approval system, audit trail, and account-disconnect controls provide a framework for supervised agent activity. Those protections are valuable, but they work best when users keep the agent’s permissions narrow and read approval prompts carefully.
Meta Muse is less suitable for tasks that require highly sensitive information, irreversible decisions, or judgment that depends on context the agent cannot reliably verify. Users should not treat an AI-generated itinerary, email, negotiation message, or purchase recommendation as proof that the underlying choice is correct. The agent can execute a workflow, while the user remains responsible for the outcome.
For most users, travel research, draft communications, and form preparation are sensible places to begin. Start with a low-cost task, inspect the audit trail afterward, and keep payment methods disconnected until you are comfortable with how Muse presents approvals. Users concerned about wider AI-agent access should also follow developments such as computer-task AI systems, since approval design and privacy controls are becoming central consumer questions across the category.
FAQ
Is Meta Muse available in the United States?
Meta Muse is available initially in the United States for people age 18 and older. Meta says the service is rolling out on iOS, Android, and muse.ai, with access through WhatsApp as well.
Can Meta Muse book travel?
Meta Muse can book travel, according to Meta’s product announcement. Users should review destinations, dates, cancellation terms, traveler details, and the final price before approving a booking.
Can Meta Muse make purchases?
Meta Muse can make purchases through Stripe Link after requesting approval for the sensitive action. Meta says Link can create one-time-use cards that hide the user’s actual card details during checkout.
Does Meta Muse use conversations to train Meta AI?
Meta Muse lets users opt out of having their interactions used to train Meta AI models, according to Meta. Users can also disconnect connected services when they no longer need the agent to access them.
Is Meta Muse free?
Meta Muse has a free tier, while Axios reported additional US subscription tiers of $20 and $100 per month. Users should confirm current plan limits and pricing before subscribing because launch details can change.