Meta Tests 'Human Concierge' for New AI Assistant Muse, Raising Employee Privacy Concerns

Meta Platforms is testing a “human concierge” feature for its new personal AI assistant, Muse, that routes some user-initiated phone calls to human contractors instead of being fully automated. The test, detailed in internal company posts seen by Reuters, has raised privacy concerns among some employees over the potential for sensitive information to be unintentionally shared.

Human-in-the-Loop Testing

The feature, referred to internally as “human agent calls,” allows the Muse AI to hand off complex phone-based tasks to a trained human agent. These tasks could include booking appointments, checking a store's inventory, or getting quotes from businesses on a user's behalf. After the call is completed, the user receives a summary and transcript within the Muse app.

According to an internal announcement, Meta enabled the human concierge test for half of its employees last week, providing an option for them to opt out. This follows an earlier phase of employee testing, or “dogfooding,” of the phone-calling capability that began in August.

Ad

Company Response and Market Context

In response to the internal discussions, a Meta spokesperson stated the test's purpose is to "get feedback so we can implement safety and privacy protections and improve features before we release them publicly." Spokesperson Daniel Roberts added that the response from employees has been "overwhelmingly positive" and that the feature would only be rolled out to the public "when it’s ready and with the proper disclosures."

The use of human oversight highlights the complexities of developing fully autonomous AI agents, a key part of CEO Mark Zuckerberg's vision for "personal superintelligence." At its launch, Meta heavily promoted Muse's advanced security, including isolated cloud computing sessions for each user. The AI assistant has seen strong initial adoption, reaching over 2.5 million downloads in the U.S. within two weeks of its debut, according to data from market intelligence firm Sensor Tower.