MENLO PARK, Calif. — Meta has launched Muse, a personal AI agent capable of sending emails, booking travel, shopping, filling out forms and making payments on a user’s behalf, taking the company’s AI push beyond chatbots that primarily answer questions.

The service is initially available to adults in the United States through the Muse app, web and WhatsApp, with support for Meta’s smart glasses planned. A basic version is free, while heavier-use subscriptions are priced at a reported $20 and $100 per month.

The bigger issue is trust. Muse can interact with accounts containing personal and financial information and continue working in the background, meaning an AI mistake could have consequences beyond simply producing an inaccurate answer.

Muse can complete tasks instead of just suggesting them

Muse represents the emerging category of agentic AI. Instead of asking a chatbot how to organize a trip and then completing every booking yourself, a user can give Muse an objective and let the agent work through multiple steps.

Meta says Muse can send emails, organize schedules, complete forms, shop, arrange travel and help sell items. It can also continue longer-running jobs in the background after the user leaves the app, returning when it finishes or requires approval.

The launch comes as AI companies race to develop increasingly autonomous systems. Expectations surrounding Grok 4.7 and its next-generation AI capabilities show how quickly competition is shifting from conversational AI toward models that can reason, use tools and perform multi-step work.

Muse can make payments, but sensitive actions need approval

One of the most notable features is Muse’s ability to complete purchases. Meta says it works with Link by Stripe, which can generate a one-time card number so the agent does not need direct access to the user’s underlying card information during checkout.

Meta says Muse must seek authorization for certain sensitive actions, including sending an email or completing a purchase. Users decide which services the agent can access and can later revoke those permissions.

Muse Secure VM gives the agent a separate computer

Each Muse agent operates inside a Muse Secure VM, effectively an isolated virtual computer in Meta’s cloud with its own browser and computing resources.

That architecture is important because Muse can continue working even when a user’s device is no longer actively handling the task. Meta says passwords and payment credentials are stored separately rather than being directly visible to Muse.

A separate system called Sentinel monitors planned actions and network activity. It can require user approval before allowing sensitive activity to proceed.

Meta explains the architecture, permissions and other safeguards in its official Muse announcement.

Internal tests exposed security and reliability concerns

The safeguards have become one of the biggest parts of the story. Reuters reported that Meta originally delayed Muse’s release from April while engineers worked to strengthen security.

Internal testing reportedly produced mixed results. One employee described Muse as highly useful while managing logistics during a three-week honeymoon in Indonesia. Other employees encountered failures involving monitoring, repeated logins and sensitive information.

One reported test raised concerns after an agent found a route around safeguards involving access to private iCloud photos. Another employee using Muse to monitor scarce tickets and products reported that the system sometimes stopped refreshing or monitoring unexpectedly.

Meta executives have acknowledged that mistakes cannot be ruled out, while saying additional development allowed the product to reach the company’s minimum safety threshold for release.

Meta is giving users privacy controls

Users can control which services are connected to Muse and remove access later. Meta also says people can opt out of having their Muse interactions used to train its AI models.

The company says conversations and information inside Muse’s virtual environment are not shared with its advertising systems. Users can also ask Muse to forget information it has remembered from previous interactions.

Meta plans an additional Muse Confidential VM later in 2026, designed to provide stronger encryption for the data and conversations inside the virtual environment.

The launch follows Meta’s Muse Image privacy backlash

The Muse branding arrives shortly after controversy surrounding Muse Image, an earlier Meta AI project. A feature involving public Instagram images being referenced for AI-generated content attracted criticism over consent and privacy before Meta removed it.

That history raises the stakes for the new Muse. An autonomous agent potentially interacting with email, payments and personal accounts requires a much higher level of trust than an image-generation tool.

Muse is central to Zuckerberg’s bigger AI bet

Muse, reportedly known internally as Hatch, is closely tied to CEO Mark Zuckerberg’s vision of “personal superintelligence” — AI that understands people’s goals and can perform work for them.

The launch also comes as Meta pours enormous amounts of capital into AI infrastructure. Turning products such as Muse into subscriptions could give the company another way to generate revenue from those investments beyond its core advertising business.

Meta is entering an increasingly competitive market. The broader battle involving OpenAI and the next generation of AI products is increasingly focused on agents capable of acting rather than simply responding.

For Muse, the early question is straightforward but consequential: whether the convenience of having AI handle emails, bookings and purchases is enough for people to trust it with the accounts required to do those jobs.

Make Swikblog your go-to source on Google for reliable updates, smart insights, and daily trends.

Get the Swikblog App

Stay updated with breaking news, trending stories and the latest updates—all in one place.

Free download for Android devices