Meta Pauses Secret Human-Assisted Calling Test for Muse AI Agent Following Internal Privacy Backlash
Meta temporarily rolled back a secret "human concierge" test for its new Muse AI assistant following internal employee complaints over privacy risks and unannounced human intervention. Third-party human contractors quietly made phone calls to local businesses on behalf of users, boosting task success rates from pure AI levels up to 95–98%. The controversy highlights how tech companies sometimes rely on undisclosed human labor to artificially inflate autonomous AI capabilities and performance metrics. It also underscores severe privacy risks when users entrust sensitive personal data to AI agents, only for that information to be exposed to human call center operators. Meta introduced human callers after businesses repeatedly hung up on automated AI voices, but employees warned that passing sessions to contractors undermined Muse's key selling point of isolated secure virtual machines. Additionally, one test call resulted in a contractor making racist comments during a bill negotiation with a service provider, prompting Meta to remove the contractor from future projects.
## BACKGROUND
Meta recently launched Muse, a personal AI agent designed to autonomously perform tasks like scheduling appointments, sending emails, and making phone calls on dedicated cloud virtual machines. Autonomous agents often struggle with phone calls due to anti-bot resistance or complex conversational edge cases, leading developers to implement human-in-the-loop fallbacks. Meta previously conducted a similar experiment a decade ago with its Messenger "M" assistant, where media reports estimated roughly 70% of tasks were actually handled by human operators.