Meta Platforms has tested a "human concierge" feature in which humans assist with the tasks of its new personal artificial intelligence (AI) assistant, Muse, Reuters reported on the 22nd, citing an internal Meta post.

Testing for the feature has currently been suspended due to concerns over data leaks and other issues.

The human concierge approach is designed to accommodate people who are reluctant to make phone calls with AI by having contract workers employed by Meta quietly handle some of the calls that Muse would otherwise make on behalf of users.

Muse is designed to make direct phone calls to third parties for tasks such as sending emails, shopping, and travel booking, as well as making hair salon appointments or checking store inventory.

A Meta spokesperson explained that the purpose of the test was to establish safety and privacy safeguards and to improve functionality.

A vice president in Meta's Superintelligence Labs division explained the rationale for introducing the human concierge by noting that success rates reached 95 to 98 percent when humans handled the calls, which was higher than when AI handled calls alone.

However, some Meta employees raised concerns that sensitive information could flow to call center contractors during the phone calls handled by Muse.

One employee wrote in an internal post, "A single bug could leak unnecessary information to a human call handler."

Another employee reported that after asking Muse to negotiate internet and cable bills, they discovered racially discriminatory remarks made by a human contract worker in the call logs.

Regarding the controversy, the vice president of Meta's Superintelligence Labs division stated, "It was a mistake to test having contractors make calls without proper notification," and announced that the feature had been suspended.

Apologizing for the racist remarks, the company stated that the personnel involved would be excluded from future Meta projects.

Reuters pointed out that this test demonstrates the reality of how tech companies rely on human labor behind the scenes to deliver on the promise of autonomous AI.

A decade ago, Meta (then Facebook) previously introduced an AI calling assistant named "M" on Messenger before shutting it down after two and a half years.

According to media reports at the time, it was estimated that about 70 percent of the tasks for that agent were actually handled by humans.

(Photo: Getty Images)