Muse's 'call on your behalf' feature exposed as partly relying on real humans; Meta has rolled back the test.

Beating AI News Flash: Just as Meta expanded the Muse "call on your behalf" test, it was revealed that some calls were actually made by humans.

Internal information obtained by 404 Media and Reuters shows that Muse can transfer tasks to human contractors at call centers, who make calls and complete requests on behalf of users. Some testers did not learn that a human had intervened until after the call ended.

Meta did this to improve the success rate. Internal tests show that with human intervention, the success rate for phone tasks can reach 95% to 98%. One practical problem is that some merchants hang up as soon as they hear the other party is an AI, and testers in the community have already encountered this situation.

Human fallback also brings privacy issues. Users originally thought their information was only being handed to AI, but in reality it may be seen by outsourced personnel. Some people inside Meta also question whether giving contractors privacy training is equivalent to a real security mechanism.

A Meta internal lead later admitted that failing to disclose in advance that humans would intervene was a mistake, and has rolled back this human takeover test. Muse's phone feature itself is still in testing, and Meta says it will add safety, privacy, and clear disclosure measures before an official launch.