Meta has temporarily rolled back a test that used human contractors to handle phone calls on behalf of its new AI assistant, Muse, following internal pushback over user privacy and safety.
The feature, dubbed a "human concierge," was designed to improve the success rates of Muse’s phone-calling capabilities by having human agents quietly jump in to complete tasks like booking haircut appointments, checking store inventory, or requesting quotes from service providers.
Why Meta Introduced Human Callers
While Muse tops U.S. app download charts with over 2.5 million downloads, its automated calling feature ran into early real-world friction.
- Robocall Resistance: Businesses frequently hung up on Muse after recognizing an automated voice.
- Higher Success Rates: Meta’s internal tests showed that human contractors could raise call completion rates to 95%–98%, compared to lower success rates for fully automated AI calls.
To address these hurdles, Meta rolled out the "human agent calls" feature to half of its employees in a test phase, enabling Muse to pass off requests to call center contractors.
Internal Pushback and Privacy Risks
The test sparked immediate concerns among Meta employees, who argued that involving call center contractors compromised user privacy and contradicted Meta's core safety promises.
- Privacy Leak Concerns: Employees pointed out that users often ask Muse to contact service providers, which could accidentally expose sensitive personal data to third-party human callers.
- Inappropriate Behavior: One employee reported that a human contractor made a racist remark during a call with a cable and internet provider to negotiate a bill.
- Undermining Security Claims: Meta launched Muse with explicit security promises—including isolating tasks in secure virtual machines—which staff felt were undercut by handing calls off to humans.
Meta's Response and Next Steps
Following employee feedback, Meta executives acknowledged that launching the test without clear disclosures was "a miss" and confirmed that the human concierge feature has been rolled back for now.
A Meta spokesperson stated that the goal of the internal test was to collect feedback and refine safety protections before any public rollout. Meta plans to continue working with merchants to improve AI calling capabilities and will only launch the feature publicly with proper disclosures once it meets safety standards.
