Meta’s much-hyped Muse AI agent is being tested with a phone-calling feature that can make restaurant reservations, book appointments and handle customer-service calls on a user’s behalf. But according to a report from 404 Media, some of those calls are not actually being handled by AI alone. Instead, Meta has added a human agent layer in which trained people can take over the call.
The discovery comes just days after Meta executives publicly announced an expansion of Muse’s outbound calling beta to US businesses. Internally, however, Meta reportedly told employees that it had added a human layer to ensure calls could be completed. The internal description said Muse could hand requests to a trained human agent, who would place the call and work through the user’s request.
Muse can make the call, but a human may be doing the talking
The distinction is important because Muse is designed to make these interactions feel automated. Meta’s internal description says Muse can call a business, handle the conversation, complete the request, and return a transcript and summary to the user. However, 404 Media reports that it is currently unclear how frequently calls are routed to human agents or when they are handled entirely by AI.
The setup has already raised questions among Meta employees, particularly around privacy. A user could, for example, ask Muse to arrange a sensitive appointment and reasonably assume that the information is being handled entirely by an AI system. Instead, a human contractor could potentially receive the request and make the call.
According to internal communications seen by 404 Media, Meta said contractors involved in the process had undergone extensive training intended to keep user data safe. Some employees questioned whether training alone was sufficient protection. Others reportedly said testers were not always told beforehand that a human had made their call.
That makes the issue bigger than simply whether Muse can successfully book a table. It touches on what users think they are handing over when they ask an AI agent to act for them.
Meta says the feature is still being tested
There is also evidence that the calling system can work without human intervention. One Meta AI researcher said Muse recently navigated a customer-service phone tree, waited on hold, spoke with a representative, and resolved the issue. Other users, however, have reported failed calls or businesses hanging up.
Meta says the feature remains in development. A company spokesperson told 404 Media that internal testing is intended to gather feedback and improve safety and privacy protections before a public rollout. Meta also said it is working with merchants and will only release the calling feature when it is ready and includes proper disclosures.
For users, that means Muse’s calling capability should currently be viewed as a beta feature rather than a finished autonomous AI service. The next step will be whether Meta can make the system reliable enough to handle calls independently, while clearly telling users when humans are involved and what information those people can access.
