Meta unveils Muse, an AI agent that can send emails and make purchases with user approval

Meta has launched Muse, a new personal AI agent that can connect to other apps, carry out tasks such as sending emails and make purchases, and keep working on longer jobs even after a user closes the app. The company says the system is built around user control, but the rollout comes as Meta continues to face scrutiny over how AI products handle sensitive personal data.

According to Meta, Muse is designed to act more like a digital helper than a simple chatbot. People can tell it what they need done, choose which apps it connects to, and decide how much access it receives. For email, Meta says users can choose whether the agent only reads messages or can also send on their behalf. The company also says the agent pauses for approval before sending an email or making a purchase.

The launch builds on Meta’s broader push to turn its AI tools into action-oriented assistants. In recent months, the company has said Meta AI can connect to email and calendar apps, create slides and handle tasks on a user’s behalf. Meta’s AI products are available through the Meta AI app, the web, WhatsApp, Instagram, Messenger and Facebook, reflecting the company’s effort to make the assistant part of its core app ecosystem.

Privacy and access controls remain central concerns

Meta says Muse runs on a dedicated virtual machine it calls Muse Secure VM, which it describes as a protected environment for both the agent and a person’s data. The company presents that setup as a safeguard for privacy, safety and security. It has also said the system can remember details that matter to a person and make suggestions proactively, which may make the assistant more useful but also increases the amount of personal information it may need to process.

Those features have raised internal concerns, according to the original reporting, because the technology could mismanage access to sensitive personal data. Meta has not publicly detailed any specific incident tied to Muse in the materials reviewed, and it is not clear whether the agent is already broadly available or being rolled out in stages. The company’s announcement focuses on product capabilities and user permissions rather than on any known failures.

The timing underscores how quickly generative AI is moving from answering questions to completing real-world tasks. For users, that could mean less friction in everyday work such as managing inboxes or arranging travel. It also means the stakes are higher: the more an AI agent can see and do, the more important its permission settings, approval steps and data protections become.