Meta tests humans to handle some calls made through its Muse AI agent

Meta has been testing a “human concierge” for its new Muse AI agent, with human contractors quietly handling some phone calls made through the digital assistant, according to internal company posts Reuters saw.

The test comes shortly after Meta introduced a phone-calling feature for Muse, its new personal AI agent that can perform tasks including sending emails, shopping and booking travel.

Some Meta employees raised privacy concerns about the arrangement, warning that involving human agents could unintentionally share sensitive information with contractors working in call centres.

Meta said it was testing the feature internally before any potential public release.

“We’re working with merchants to continue improving this potential calling feature, and will only roll it out when it’s ready and with the proper disclosures,” Meta spokesperson Daniel Roberts told Reuters.

Muse’s phone feature allows users to ask the AI agent to call US businesses on their behalf.

It can handle tasks such as booking a haircut, checking whether a product is available in a shop or requesting quotes from contractors.

After completing a call, Muse provides the user with a transcript and summary of the conversation.

Meta began having employees test the phone-calling capability in August, then gradually made the feature available to Muse users after the app’s launch.

But internally, Meta also began testing another layer: humans.

The “human concierge” feature, also called “human agent calls”, allowed Muse to hand a request to a trained human agent who would place the call and complete the task.

Meta enabled the feature for about half of its employees during internal testing. Employees who did not want to participate could opt out.

Some employees questioned what would happen when a call involved sensitive information.

A Muse user might, for example, ask the agent to make a booking or carry out another task requiring personal details without realising that a human contractor could become involved.

Meta said feedback from employees testing the feature had been “overwhelmingly positive”.

The company said the purpose of testing was to identify issues and introduce appropriate safety and privacy protections before deciding whether to release the feature publicly.

The human-concierge system therefore should not be confused with a hidden feature currently being used across all public Muse calls.

The disclosure comes as Muse enjoys a strong start following its September 8 launch.

Market intelligence firm Sensor Tower said the app had recorded more than 2.5 million downloads and topped US app-download charts within two weeks of its debut.

Separate Apptopia data previously cited by Reuters put downloads in the US and Canada at 2.8 million during its first 12 days.

Muse differs from conventional AI chatbots because it is designed to perform tasks rather than answer questions.

It can browse websites, fill out forms, send emails, make purchases and arrange travel on a user’s behalf.

Meta says Muse operates inside a dedicated secure virtual machine and keeps sensitive information such as passwords in separate secure storage.

The company has heavily promoted privacy and security as central features of the product.

That makes the human-concierge experiment particularly significant: it highlights the additional privacy questions that can arise when an AI agent hands part of a user’s task to a person.

Meta says it would roll out publicly only when it is ready and with appropriate disclosures.