Meta has been testing human contractors to handle some phone calls placed through its personal AI agent Muse, according to a Reuters report published September 22. The test involved contractors quietly handling calls placed through Muse, according to internal company posts Reuters accessed.
Muse is designed to perform tasks such as sending emails, shopping and booking travel. Its phone-calling feature lets users ask the agent to contact US businesses to book appointments, check whether products are in stock or obtain contractor quotes.
The human-contractor test came shortly after Meta launched the calling feature. Internal posts also showed that some users had encountered problems with AI-generated calls, including one case in which an insurance company ended a call after recognising the AI system.
Meta enabled the human-concierge feature for half of its employees last week, according to an internal post cited by Reuters. Under the test, Muse could hand a request to a trained human agent, who would place and handle the call. Employees who did not want to participate could opt out.
Some Meta employees raised concerns that human agents could be exposed to sensitive information during calls. Internal posts cited by Reuters showed employees warning that information could potentially be shared unintentionally with contractors working in call centres.
One employee also reported that a human contractor made a racist reference during a call with an internet and cable provider. Meta’s vice president in its Superintelligence Labs acknowledged that starting contractor-placed calls without proper disclosures was a mistake and said the company had “rolled back this feature” for now.
The executive also said some tests showed human-handled calls achieving success rates of 95% to 98%, compared with lower success rates for AI calling.
Meta spokesperson Daniel Roberts said the company was working with merchants to improve the potential calling feature and would roll it out when it was ready and with the proper disclosures.
Muse has recorded more than 2.5 million downloads since its release, according to Sensor Tower data cited by Reuters. Meta has also highlighted security measures for the agent, including dedicated secure virtual machines and separate storage for sensitive information such as passwords.
- Anthropic Posts $17 Bn in Q3 Revenue, Swings to Operating Loss
- OpenAI Signs Four Agreements With Abu Dhabi Across Government, Energy And Investment
- Deep-Tech and Green Finance to Drive India’s Climate Investments in 2027
- Google’s Project Suncatcher Satellite Generates First AI Response in Orbit
- DeepSeek Proposes Super-Agent Architecture to Automate AI Coding Workflows
- Arrowhead AI Launches Fletch Voice Model for Multilingual Customer Calls

