Meta's Muse AI phone calls are secretly made by human workers

 

Meta's Muse AI phone calls are secretly made by human workers

When Meta executives announced last month that its AI agent Muse could make phone calls on your behalf, booking restaurant tables, scheduling haircut appointments, they framed it as a breakthrough in autonomous AI capability. What they did not mention publicly, but did tell employees internally, is that the feature relies on human workers in call centers to place and handle those calls. According to internal communications obtained by 404 Media, Meta quietly added what it described as "a human agent layer" to complete those calls, and early testers were not informed that a person, rather than an AI, was making the call until after it had already happened. The revelation touches on serious privacy concerns that Meta's own employees flagged loudly on internal message boards.

When a user asks Muse to schedule a sensitive medical appointment, for example, that request and the personal details attached to it may be routed to a human contractor, a fact users had no way of knowing. Meta told employees that contractors received extensive training to keep data safe, but workers pushed back, with one noting bluntly that training "is not a security mechanism." Several employees warned that launching the feature without clear disclosure would generate damaging press and destroy the goodwill the product had earned among early adopters. Meta told 404 Media that internal testing is a standard part of its development process and that the company will only roll out the calling feature publicly "when it's ready and with the proper disclosures," suggesting the human-assisted version of Muse calls is not the final product, though exactly how and when the handoff between AI and human agent occurs remains unclear.