Meta is testing a “human concierge” approach for its Muse personal AI assistant by having human contractors quietly handle some phone calls placed through the service, according to internal company posts reviewed by Reuters and echoed by multiple outlets. The test follows Meta’s recent public launch of a Muse phone-calling feature that lets the agent place calls to US businesses to carry out tasks such as checking stock, booking appointments, or requesting quotes.
Several reports say Meta began the internal testing shortly before and around the feature’s rollout to users, including enabling the contractor-placed calls for a portion of employees. Company officials acknowledge employee concerns and say the purpose of the trial is to gather feedback to improve safety and privacy protections before wider release. Employees who raised objections cite risks that sensitive information could be shared with outside call-center contractors, including concerns about whether disclosures are adequate.
Outlet accounts also note that Meta’s vice president in its Superintelligence Labs unit describes the initial approach as a “miss,” and reports that Meta has rolled back the contractor-placed calling feature for now. The articles frame the differing emphasis as: improving call success rates versus addressing privacy, security, and disclosure concerns. Some reports also mention that Meta previously highlighted isolation and secure handling for Muse sessions, which critics say could be undermined by introducing human callers.
