Reuters reports that Meta has been testing a 'human concierge' for Muse, its new personal AI agent, where human contractors handle some of the phone calls Muse places for users. The report is based on internal company posts seen by Reuters. Muse launched on September 8, and its calling feature can phone US businesses to book a haircut, check whether a store has an item in stock, or collect quotes.

Apparently, not every call went smoothly, with one employee saying their insurance company kept hanging up once it realized it was talking to an AI. Meta reportedly enabled the human option for half of its employees last week, with an opt-out group for anyone who didn't want in. It looks like that didn't go over well internally.

Some employees raised privacy concerns, warning that sensitive information could accidentally end up with call center contractors. One of them reportedly wrote, 'We are one bug away from unnecessary information being leaked to human callers.' It gets even messier as another employee, who had Muse call his internet and cable provider to negotiate his bill, said the transcript showed the contractor made a racist reference. A Meta Superintelligence Labs vice president reportedly acknowledged it was a mistake to start the test without proper disclosures and said the feature has been rolled back for now.

Still, they're not completely backing out. A Meta spokesperson reportedly said employee feedback has been overwhelmingly positive and that the calling feature will only roll out once it's ready, with the appropriate disclosures. The report mentions only employees, so we don't know whether regular users' calls were ever routed to contractors.

Given Meta's history, it will be difficult to trust them with proper disclosures and handling of private data, especially when requests are being routed through humans.