Meta has been experimenting with a 'human concierge' service for its newly rolled-out personal AI assistant, Muse. This service, which was revealed to employees shortly after Muse unveiled its call-making capability, entails human contractors managing some of the calls made by Muse behind the scenes. Within just two weeks of its launch, Muse soared to the top of the U.S. app download charts, boasting features such as autonomous email drafting, shopping, travel booking, and more. The call feature enables users to assign tasks like scheduling appointments, inquiring about products, and requesting quotes to merchants. Yet, this trial has sparked privacy concerns among employees. One staff member recounted enduring racist comments from a human contractor while using Muse to call their carrier for bill negotiation. The vice president linked to Meta's superintelligence lab admitted that initiating the trial without adequate disclosure was an oversight and has temporarily suspended the feature. He highlighted that human-assisted calls could boost call success rates to 95-98%, extended apologies to the impacted employees, and permanently deactivated the involved human agent. This action echoes Meta's prior experiment with its digital assistant M on Messenger, where roughly 70% of tasks were fulfilled by humans. Meta clarified that gathering test feedback is intended to enhance safety, privacy safeguards, and functionality, and the call feature will only be officially rolled out once it's fully prepared and properly disclosed. Since its debut, Muse has been downloaded over 2.5 million times, propelling Meta's stock price up by a cumulative total of more than 20% and augmenting its market value by over $200 billion. Nevertheless, on the day the trial was disclosed, the company's stock price experienced a slight dip of 0.6%. Some internal staff members noted that the 'human concierge' model could potentially compromise Muse's core security selling points, such as isolated conversations and encrypted storage of sensitive data. Previously, calls powered solely by AI encountered problems where recipients would hang up once they realized it was an AI. To tackle this, Meta commenced testing human-agent calls with half of its employees last week and offered employees the choice to opt out.