Meta's AI Calls Secretly Used Humans. One Made a Racist Remark

Meta launched a new AI assistant called Muse on September 8, and it quickly became the most downloaded app on iPhones in the U.S. One of its big selling points is a beta feature that lets the AI make phone calls on your behalf — booking a haircut, checking if a store has something in stock, or getting quotes from contractors. It sounds effortless. But according to a report from Reuters, some of those calls weren't made by the AI at all.

When the AI voice couldn't handle a task, Meta quietly had a real human step in. Why? Because the AI wasn't good enough on its own. One user said his insurance company kept hanging up as soon as the bot admitted it wasn't a person. Internally, Meta apparently found that humans pushed the call success rate into the high 90% range, compared to a "lower percentage" for the AI alone. Some employees worried this could leak your private details to a stranger — the difference between a human calling your barber and a human calling your urologist.

Then came the bigger problem. One employee who reviewed a call transcript claimed the human operator had made a racist reference. A Meta vice president apologized, said the company had stopped working with that individual, and described the whole approach as "a miss." Meta says it has rolled back this feature.

A Meta spokesperson told reporters that internal testing is normal and helps the company add safety and privacy protections before a public launch. It's still unclear whether all AI calls were paused or only the ones involving human helpers. Either way, it's a reminder that when AI falters, the fallback is often a person — and you may never know which one you got.