NEW YORK CITY, New York: Meta has temporarily rolled back a test that used human contractors to handle some phone calls requested through its new Muse AI assistant after employees raised concerns about privacy and disclosure.

The Facebook and Instagram owner began testing the "human concierge" shortly after publicly launching Muse's phone-calling feature, according to internal company posts seen by Reuters. The arrangement allowed some requests made through the digital assistant to be passed to human contractors who would make the calls.

Muse, launched earlier this month, is designed to carry out tasks such as sending emails, shopping, and booking travel. Its calling feature allows users to ask the agent to contact U.S. businesses to book haircuts, check whether products are in stock, or obtain quotes from contractors.

Some Meta employees warned that involving human contractors could inadvertently expose sensitive information.

"It's baffling to me why we think this feature is worth the risk," one employee wrote in an internal post. "We are one bug away from unnecessary information being leaked to human callers."

Another employee said he had asked Muse to call an internet and cable provider to negotiate his bill and later found in a transcript that the human contractor had made a racist reference during the call.

A vice president in Meta's Superintelligence Labs unit acknowledged internally that "it was a miss" to begin testing contractor-placed calls without appropriate disclosures and said Meta had "rolled back this feature" for now.

The executive said Meta was still "hillclimbing AI calling," adding that tests suggested human involvement could raise the success rate of calls to between 95 percent and 98 percent, compared with the "lower percentage of AI calling."

Meta spokesperson Daniel Roberts said employee response to the test had been "overwhelmingly positive" and that it was intended to gather feedback before any public release.

"We're working with merchants to continue improving this potential calling feature, and will only roll it out when it's ready and with the proper disclosures," Roberts said.

The test comes as Muse gains rapid traction. The app has recorded more than 2.5 million downloads since its release, according to data from market intelligence firm Sensor Tower. Meta's stock has risen more than 20 percent since the launch, adding more than US$200 billion in market value and reaching its highest level in more than seven months as of Monday's close.

Meta emphasized Muse's security protections at launch, including dedicated cloud-based virtual machines to isolate individual users' sessions and separate storage for sensitive information such as passwords. Employees questioning the human concierge test cited those safeguards in raising concerns.

Meta began employee testing of Muse's calling capability in August, then gradually made it available to users after the app debuted. Internal posts showed that some calls encountered resistance when businesses realized they were dealing with AI.

Meta subsequently enabled the human concierge feature for half of its employees last week. Employees who did not want to participate could opt out, according to the internal posts.