Meta is testing a "human concierge" feature for its new Muse AI assistant, using human contractors to handle phone calls. The test triggered internal privacy concerns among employees worried about sensitive data reaching call centers. A Meta SuperIntelligence Labs executive admitted launching the test without proper disclosures "was a miss," prompting Meta to roll back the feature. Despite the setback, Muse has topped US charts with over 2.5 million downloads.
Muse human concierge testing
- ▪Meta is testing a "human concierge" feature for its new personal AI assistant, Muse, where human contractors handle some phone calls placed via the digital agent
- ▪Meta enabled the human concierge feature for its Muse AI assistant, also referred to as "human agent calls," for half of its employees in September 2026 shortly after Muse's public phone-calling feature launched
Feature rollback after disclosure miss
- ▪Meta rolled back the human concierge calling feature for Muse following employee concerns over the lack of proper disclosures about testing contractor-placed calls for Muse's phone-calling feature
- ▪A vice president in Meta's SuperIntelligence Labs unit acknowledged in an internal post that starting the contractor-placed calls for Muse's phone-calling feature without proper disclosures "was a miss."
AI versus human calling success
- ▪Internal Meta posts indicated that having human contractors make phone calls placed via Meta's Muse AI assistant, instead of having Muse's AI system make the calls, could increase the call success rate to the 95% to 98% range
- ▪A decade prior to Meta's Muse AI personal assistant, Meta trialed a digital assistant called "M" in Messenger where media reports estimated around 70% of tasks were carried out by humans
Muse app download leadership
- ▪Meta CEO Mark Zuckerberg described Meta's Muse personal AI assistant as his vision for supplying "personal superintelligence" to billions of daily users
- ▪According to Sensor Tower data, Meta's Muse app topped US download charts with more than 2.5 million downloads in the two weeks following its debut
Meta safety protection measures
- ▪Meta designed Muse to run each user session on an isolated virtual machine in the cloud to keep individual user data secure
- ▪Meta designed Muse to store sensitive user information, such as passwords, in separate secure storage to maintain security boundaries
Debatable claims
- ▪Meta should use human contractors to improve Muse AI call success rates
- ▪Meta must obtain explicit user consent before routing Muse AI calls to human contractors
Story comments
Loading comments…