Meta is testing a new feature for its Muse AI agent that allows it to make phone calls on behalf of users to services like booking appointments. The company recently utilized contractors to handle these calls during the testing phase, granting them access to the conversation content.
Internal reports indicate that contractors achieved a success rate of 95 to 98 percent in completing tasks, outperforming the automated voice bot. However, the involvement of human workers raised privacy concerns among Meta employees, particularly regarding the potential exposure of sensitive user data.
Meta confirmed that the use of contractors for these calls was an error and has since ceased the practice. The company stated that a contractor who made racist remarks during a call has been permanently banned from working on any Meta projects.
This is not the first time Meta has relied on human labor to support AI features. A decade ago, the company's M assistant for Messenger also relied on humans to handle approximately 70 percent of its tasks. Meta claims these experiments help identify security and privacy vulnerabilities.
Since its launch this month, the Muse application has been downloaded over 2.5 million times. While the company initially promised that each agent would operate within an isolated virtual machine, the recent testing phase involved human intervention to accelerate the identification of system weaknesses.
