Meta’s Muse AI call feature secretly relies on humans

Meta’s new “personal assistant” AI agent Muse is using good old fashioned human beings in call centers to handle phone calls to external businesses, it has emerged.

The company recently expanded an outbound calling feature for beta testing that allows Muse to make calls to US businesses, including making restaurant reservations and booking appointments.

According to Muse's principal engineer Ryan Fox, “phone calling was one of the top user requests, but users would likely assume the process was entirely automated, not handled by a human agent."

However, an internal Meta memo, obtained by Reuters and 404 Media reveals that the company has also introduced what it calls a “human agent layer” to help complete calls.

"Muse doesn't just dial a number. It calls a business on your behalf, handles the conversation, completes your request, and reports back with a transcript and a summary,” the memo reads.

"It also no longer works alone. Muse is now able to hand requests to a trained human agent, who places the call and works it through."

Meta doesn’t appear confident that people outside Meta would view this “human layer” as a perk however, and the memo reminds staff that Muse is a "confidential, pre-launch product.”

Please don't share [Muse], or anything it produces, with anyone outside the company,

– warns Meta staff memo, as seen by 404 Media.

It is unclear how frequently calls are transferred to humans rather than completed by AI.

Privacy concerns

The arrangement appears to have stoked privacy concerns among Meta’s staff, because users who are testing the feature were apparently not told that they would be providing information to a person rather than the AI.

According to a 404 Media report, one Meta employee said that a person testing the calling feature was only informed after the call had been completed that it had been handled by a human.

The employee added that handing part of the AI’s task over to humans without full disclosure, even in Beta, was a “bad, bad idea.”

Another told Reuters, "It's baffling to me why we think this feature is worth the risk.

We are one bug away from unnecessary information being leaked to human callers,

warns Meta employee in memo seen by Reuters.

Racist comment

Another Meta employee, who tested Muse by asking it to call an internet and cable provider to negotiate his bill, said a transcript showed the human contractor had made a racist reference during the call.

Responding to these reports, the vice president in Meta's Superintelligence Labs unit told Reuters that "it was amiss" to start testing contractor-placed calls without proper disclosures and said the company had "rolled back this feature" for now.

Meta is still hillclimbing AI calling,

– says Meta spokesperson

Meta was still "hillclimbing AI calling," she added, noting that some tests indicated having humans make the calls could get their success rate up to the 95% to 98% range, instead of the "lower percentage of AI calling."

She also apologized to the employee who reported the mention of race during a Muse call and promised that the "human agent" involved would not work on any Meta projects going forward.

Banished from Amazon

The Muse agent, which has topped US app download charts over the past 2 weeks since its debut, is designed to autonomously perform tasks such as sending emails, shopping, and booking travel on a person's behalf.

However, shoppers who want to use Muse to buy things on Amazon will have to look elsewhere, after it was reported on Monday that the retail giant has requested that Meta remove the Muse agent.

Amazon’s website rules block many AI assistants, including ChatGPT, Claude, Copilot, Perplexity, Grok, and Google crawlers.