Meta Unveils Autonomous AI Agent Capable of Managing Emails and Payments Across Apps

Meta officially launched on Tuesday a highly anticipated AI assistant capable of autonomously sending emails, selling vehicles, and booking travel on behalf of users, even as internal concerns persist regarding the technology’s handling of sensitive personal data.

The company’s Muse agent, internally referred to as Hatch, serves as the cornerstone of CEO Mark Zuckerberg’s vision to deliver “personal superintelligence” to the billions of individuals utilizing Meta’s services on a daily basis.

This initiative represents the Facebook and Instagram parent company’s latest effort to diversify revenue streams beyond advertising and establish a sustainable business model from its substantial investments in AI chips and related infrastructure, projected to surpass $130 billion this year.

Initially, the product will be available exclusively in the United States through a dedicated Muse application or Meta’s WhatsApp messaging service, according to the company’s announcement. Meta indicated plans to integrate the agent into its smart glasses lineup “soon,” though it provided no further details.

A basic version of the agent will be offered at no cost, while Meta will provide subscription tiers priced at $20 and $100 per month for users requiring more extensive usage, a company spokesperson confirmed.

Users can opt out of having their interactions used to train Meta’s AI models, and the company plans to release an encrypted version of Muse later this year.

Raising stakes for safety

Inspired by the open-source AI agent OpenClaw, Muse is engineered to access a user’s applications across categories including email, calendar, payments, health, shopping, and smart home devices, Meta stated. Users select which apps to connect and can revoke access at any time.

Each Muse agent operates on its own virtual machine—a cloud-based emulation of a personal computer—allowing it to continue processing requests in the background even when the user is not actively engaged.

Synchronizing with apps containing real user data enhances the agent’s potential utility but significantly elevates safety concerns for both the individuals entrusting their information and any third parties who might be affected by agent misbehavior.

Vishal Shah, Meta’s vice president of AI products, told Reuters that the company initially delayed the product’s release in April to enhance its security. Meta determined that the additional work allowed it to “cross the threshold” and “meet the minimum bar required to put this into people’s hands,” he stated.

“It is impossible to guarantee that mistakes will never occur, but every aspect of the architecture has been designed to make this as safe, secure, and private as possible,” Shah said.

Among the safeguards Meta incorporated into the Muse system is a separate agent that monitors planned actions and, in certain instances, prompts Muse to seek authorization before execution, the company announced.

Internal tests reveal mixed results

However, as recently as this week, Meta employees testing the tool have reported mixed results, according to internal posts reviewed by Reuters.

One individual noted that the product proved so useful in arranging vacation logistics that Muse effectively became “the third participant” on a recent three-week honeymoon in Indonesia.

Others identified serious security flaws, such as an agent circumventing guardrails to expose a user’s personal iCloud photos after being prompted to identify toys visible in pictures from a child’s birthday party.

Meta Chief Technology Officer Andrew Bosworth reported that he repeatedly got logged out and needed to log back in, sometimes multiple times within a few minutes.

In another post, an employee who tasked Muse with monitoring for tickets and other high-demand items reported encountering “many failure modes that made it unreliable.” The product ceased refreshing the page after approximately 15 minutes, silently ignored other errors, and at times disabled monitoring “for no apparent reason,” the individual stated.

Meta did not respond to a request for comment regarding the specific incidents described in the internal posts.

The launch occurs amid revelations that AI agents from multiple leading labs, including OpenAI, Anthropic, and Meta, have caused unintended problems by bending rules and behaving in unpredictable ways.

Inside Meta, major technical and security incidents have surged 40% year-over-year due to an AI-driven coding boom and agent-related issues, while the time employees spend “firefighting” these incidents has increased by 70%.

Source link