Automated Muse tool arrives in United States to run tasks By Luís Henrique Costa
2 min read
 |
Meta deployed its artificial intelligence assistant Muse across the United States on September 8, 2026, to execute operational tasks including sending emails, purchasing tickets, and booking lodging without manual oversight. The software operates through a dedicated application alongside WhatsApp, while company representatives confirmed plans to expand the tool to smart glasses later.
Direct account control across consumer services
Muse acts on behalf of account holders by logging into accounts connected to banks, appointment calendars, health trackers, digital stores, and smart home hardware. Users limit which programs the assistant inspects through options in the settings dashboard. Account holders can revoke these access permissions at any time.
Engineers built the tool from OpenClaw, an open model circulated early this year under internal code name Hatch to advance Mark Zuckerberg’s automation roadmap for Meta users.
Security adjustments delayed the initial launch schedule
Vishal Shah, vice president of artificial intelligence at Meta, confirmed that executives postponed the release from April to improve security parameters, personal privacy, and processing reliability. Shah defended the multi-layered system protections, noting that ‘it is impossible to state that there will never be an error, but every part of the architecture was designed to make the product as safe, secure, and private as possible.’
Internal trial records split Meta personnel.
- One employee noted that Muse arranged travel schedules so accurately that it functioned like a third participant during a honeymoon.
- Another tester watched the program freeze after 15 minutes during a competitive ticket sale, halting page tracking without issuing an alert.
- System audits revealed a security breach where the program accessed private iCloud photos while searching for toys in a child’s birthday celebration album.
Executive testing logs persistent software crashes
Andrew Bosworth, chief technology officer at Meta, documented repeated crashes during personal tests. These persistent failures forced Bosworth to re-enter passwords every few minutes. OpenClaw had previously generated public concern over unintentional data leaks in daily routines.
Meta did not provide comment. Company leadership left formal inquiries regarding the internal test failures unanswered.
