NEW YORK: Meta has been testing a “human concierge” service for its personal AI assistant, Muse, allowing human contractors to handle some phone calls placed by the digital agent, according to internal company posts reviewed by Reuters.
The Facebook and Instagram owner informed employees about the test last week. The move came shortly after Meta publicly launched Muse’s phone-calling feature.
Muse is designed to perform tasks on behalf of users. These include sending emails, shopping and booking travel.
The phone feature allows users to ask Muse to call US businesses and handle routine tasks. The AI can book appointments, check product availability and request quotes from contractors.
Employees Raise Privacy Concerns
Some Meta employees raised concerns about human contractors handling calls, according to the internal posts.
They warned that sensitive information could unintentionally reach call-centre workers during these interactions.
One employee questioned the risks of the feature and warned that a technical error could lead to unnecessary information being shared with human callers.
Another employee said he asked Muse to call an internet and cable provider to negotiate his bill. He later found that a human contractor had made a racist reference during the call.
A vice president in Meta’s Superintelligence Labs acknowledged that the company made a mistake by testing contractor-placed calls without proper disclosures.
She said Meta had temporarily rolled back the feature.
However, she added that the company would continue improving its AI calling technology. Internal tests showed that human contractors could raise the success rate of calls to between 95% and 98%, compared with a lower success rate for AI-only calls.
The executive also apologised to the employee who reported the racist remark. She said the contractor involved would not work on any Meta projects in the future.
Meta Says Test Was Designed to Improve Safety
The test highlights the role of human workers behind some AI systems that promise autonomous services.
Meta previously conducted a similar experiment with its digital assistant M on Messenger. Media reports later estimated that humans handled around 70% of the assistant’s tasks.
A Meta spokesperson said employee feedback on the Muse test had been “overwhelmingly positive”.
The spokesperson said the company wanted to gather feedback before publicly releasing the feature and use it to improve safety and privacy protections.
Meta said it was also working with merchants to improve the calling feature. The company will only roll it out more widely after it is ready and includes appropriate disclosures.
Muse Gains Millions of Downloads
Muse has recorded more than 2.5 million downloads since its launch, according to data from market intelligence firm Sensor Tower.
The AI assistant has also gained significant attention in the US since its debut.
Meta began employee testing of the phone-calling feature in August. The company then gradually expanded access to Muse users after the app launched.
Some users have reported problems with AI-generated calls. One employee said an insurance company repeatedly ended calls after hearing that Muse was an AI system.
Meta enabled the human concierge feature for about half of its employees last week to address such problems.
The feature, also known internally as “human agent calls”, allows Muse to transfer a request to a trained human agent. The agent then places the call and handles the task.
Employees who did not want to participate could opt out, according to the internal announcement.
Human Calls Raise Questions About AI Privacy
Meta has promoted security and privacy protections as key features of Muse.
The company said each AI agent operates through its own secure virtual machine. The cloud-based system keeps user sessions isolated and stores sensitive information, such as passwords, separately.
Employees who criticised the human concierge test said human involvement could weaken those protections.
The test therefore highlights a broader challenge for AI companies. They must balance autonomous AI systems with the practical need for human assistance when automated agents struggle to complete tasks.
Meta has said it will continue improving Muse’s calling technology before deciding on a wider ro
