Meta quietly tested having human contractors handle some phone calls made by its Muse AI agent, according to internal company posts seen by Reuters. Callers were not told a person was behind the 'AI' on the line.

The 'human concierge' test was switched on for half of Meta's employees, with an opt-out, and covered calls to US businesses such as booking haircuts or getting contractor quotes. Internal tests put human-placed calls at a 95 to 98 percent success rate, far above AI-only calls, a Meta spokesperson said.

Employees raised privacy concerns, warning that sensitive details could reach call-centre contractors. One employee said a transcript showed a contractor made a racist reference on a call, and Meta apologized, saying the person would not work on Meta projects again.

A vice president in Meta's Superintelligence Labs admitted it was a 'miss' to start the test without proper disclosures, and the company said it had rolled the feature back for now. It added that a public release would come with 'appropriate disclosures and safeguards.'

The episode highlights a wider transparency gap in AI customer service: if users cannot tell when a human takes over, trust in the whole system suffers.