Meta Tests Human-Handled Calls in Its Muse AI Agent

Summary

Meta is testing a phone feature for its Muse AI agent that can call U.S. businesses to handle tasks such as restaurant reservations and haircut appointments. Although Meta publicly describes Muse as making the calls, internal communications reviewed by 404 Media say the system now includes a “human agent layer” that can hand requests to trained people in call centers. Those agents may place the call, complete the request, and return a transcript and summary, but Meta has not disclosed how often calls are routed to humans or how users are informed beforehand. The feature remains a confidential, pre-launch product being tested internally through company dogfooding. Employees raised concerns that users could share sensitive information, including details related to medical appointments, while assuming that only a secure AI system was involved. They also questioned whether contractor training was an adequate security mechanism and warned that undisclosed human involvement could create privacy risks and negative publicity. Some testers reportedly learned only after a call that a human had made it, while other internal users described both successful and unsuccessful calls. Meta said employee testing is intended to improve safety and privacy protections, and that the feature will be released only when ready and with proper disclosures.