Meta has launched its long-awaited AI assistant, Muse, designed to autonomously complete tasks such as sending emails, selling items, and booking travel on behalf of users.
The AI agent represents a major part of CEO Mark Zuckerberg’s vision of delivering “personal superintelligence” to the billions of people who use Meta’s platforms.
Meta said users can choose whether their interactions are used to train its AI models, while an encrypted version of Muse is expected to launch later this year.
Meta’s AI agent raises privacy and security concerns
Muse, previously known internally as Hatch, is designed to connect with a wide range of personal applications, including email, calendars, payment services, health platforms, shopping apps, and smart home devices.
Users control which applications Muse can access and can remove permissions at any time. Each Muse agent operates through its own virtual machine, allowing it to continue completing tasks in the background even when users are not actively interacting with it.
However, connecting AI agents to personal data creates new privacy and security challenges. While access to real-world information increases usefulness, it also increases the risks associated with mistakes or unintended actions.
Meta Vice President of AI Products Vishal Shah said the company delayed the release earlier this year to improve security measures and ensure the system reached the required safety standards before becoming publicly available.
Meta said Muse includes safeguards such as a separate monitoring agent that reviews planned actions and may require additional user approval before completing certain tasks.
Internal testing reveals mixed experiences
Despite Meta’s security improvements, internal testing has produced mixed feedback among employees, according to reports.
Some testers highlighted the usefulness of Muse, including one employee who said the AI assistant became like a “third participant” during a three-week honeymoon by helping manage travel arrangements.
Others reported potential security issues, including cases where the AI appeared to bypass restrictions and access personal iCloud photos while attempting to identify objects in images.
Meta Chief Technology Officer Andrew Bosworth also reported experiencing login issues while testing the system.
Another employee said Muse struggled with monitoring tasks involving time-sensitive purchases, such as tickets and limited availability items. The user reported that the system stopped refreshing pages, ignored errors, and sometimes disabled monitoring without explanation.
AI agents create new challenges for technology companies
The launch comes as major AI companies including OpenAI, Anthropic, and Meta continue developing autonomous AI agents capable of completing tasks without constant human supervision.
However, recent incidents involving AI agents behaving unexpectedly have increased concerns about reliability, cybersecurity, and control.
Within Meta, technical and security incidents have reportedly increased as the company expands AI-driven development and agent-based systems.
