Meta Testing Human Callers for AI Agent Muse
According to an exclusive report by Reuters (Tuesday), citing internal company postings, Meta has been testing a "human concierge" for its Muse AI agent, with contractors taking over some of the calls placed through the agent.
Last week, Meta informed staff about this trial—after launching the calling feature for Muse. For half of Meta employees, the 'human agent calls' option was enabled (opt-out available).
"Muse is now able to hand requests to a trained agent, who places the call and works it through," stated a Meta spokesperson, per Reuters.
Some employees expressed concerns about sensitive information potentially reaching call center contractors. Daniel Roberts, a Meta spokesperson, confirmed employee feedback was "overwhelmingly positive" and that the test aimed to shape safety and privacy protections before public release.
He added: "We will only roll it out when it's ready and with proper disclosures," though he didn't specify what those would be or who would receive them.
Launched September 8 in the U.S., Muse acts as an agent that sends emails, makes purchases, and books travel on users' behalves. Its calling feature forwards requests to phone businesses—for scheduling haircuts, checking stock, etc.—and then provides a transcript. The app has reached 2.5 million downloads (according to Sensor Tower data).
At launch, Meta emphasized privacy: each agent operates in its own virtual machine. However, these statements refer to software location rather than human interaction.
The company's previous attempt at this involved Facebook’s M, a Messenger assistant supported by human "trainers." It never progressed beyond limited beta testing and was shut down in 2018 (achieving only about 30% automation).
Meta positions Muse as an agent who works around the clock on users' behalf, with human assistance on some calls.
By Ana-Maria Stanciuc, Editor-in-Chief, TNW
(Note: All italicized text, quoted content, and bolded text preserved from original.)