Meta has taken a new step in the evolution of artificial intelligence with the launch of Muse, a personal agent designed not only to answer questions but to execute tasks on behalf of the user.
The company presented it on September 8 and began its rollout in the United States for iOS, Android, and muse.ai, with future integration into its AI glasses. It can also be used directly from WhatsApp.
The technological shift lies in the architecture. Muse can send emails, book travel, complete forms, organize tasks, and make purchases while continuing to work even when the user closes the application. To do this, it uses Muse Spark, Meta’s model developed specifically for agentic operations, on an infrastructure called Muse Secure VM.
The difference from a conventional chatbot is fundamental: the system does not finish its work when it generates a response. It can interpret an objective, use tools, maintain context, and execute a sequence of actions.
Muse Incorporates Its Own Cloud Computer
Meta designed Muse Secure VM as a dedicated virtual machine where the agent and data associated with the services the user connects operate. The environment incorporates its own browser and remains isolated from the environments corresponding to other users.
This architecture resolves a problem that arises when an AI acquires action capability. A model that only generates text can make a mistake in a response; an agent that has access to email, e-commerce, or booking systems can turn an error into a real action.
That is why Meta separated execution and authorization functions through Sentinel, an independent process that supervises Muse’s activities. The company establishes that no agent action can reach the Internet without passing through this control layer.
The user can also decide which applications Muse has access to and what operations it can perform. In email, for example, it can be limited to reading or allowed to send messages on the user’s behalf. The system also maintains a record of actions performed and planned.
Autonomy Introduces a New Surface of Risk
The ability to act makes security a property of the entire architecture and not only of the artificial intelligence model.
Muse can use securely stored credentials without the agent having direct visibility over passwords or payment methods. For purchases, Meta integrates Stripe’s Link payment system and uses single-use virtual card numbers.
Meta also states that conversations and data stored in the virtual machine are not shared with its advertising systems and that users can exclude their interactions from AI model training.
The company is also preparing Muse Confidential VM, scheduled for late 2026. In that architecture, data and conversations will be encrypted using a key that will remain under the user’s exclusive control, so that not even Meta can access the protected content.
The challenge, however, goes beyond encryption. Internal testing reported by Reuters identified security and reliability issues that contributed to delaying Muse’s original launch. The issue demonstrates an inherent difficulty of agentic AI: the greater its autonomy, the greater the need to control its permissions, context, and actions.
From Chatbot to Autonomous Digital Operator
The launch of Muse marks a technological transition that may have greater importance than any of its individual functions.
This scheme transforms the relationship between user and software. Instead of opening five applications and manually executing a sequence of operations, the user can express an objective and allow the agent to determine how to achieve it.
The evolution also explains why Meta considers it important for Muse to operate in the background and retain memory about user preferences. The objective is not only to answer a specific query but to maintain goal-oriented activity over longer periods.
The company is taking this model toward mass distribution: Muse is free for most of its functions and will have subscription plans for those who need additional use. Its future arrival on AI glasses could further expand the concept of a persistent agent integrated into everyday life.
SOURCE and PHOTO: https://es.investing.com/
