Meta used human contractors to handle some Muse AI phone calls

By Splitfeed EditorialLast updated Sep 26, 2026, 19:17 UTC

During an internal early test of its Muse AI assistant, Meta routed some phone calls to human contractors rather than handling them entirely through AI. The arrangement came to light via leaks reported around September 22 and 23, 2026, just ahead of Meta's Connect keynote event, where Muse had been positioned as an autonomous AI agent capable of making calls on a user's behalf. Gizmodo reported that at least one contractor made racist remarks during test calls, adding a conduct dimension to the disclosure. Benzinga separately noted that human callers outperformed AI alone in the internal tests, suggesting the hybrid model may have reflected a performance gap rather than a purely strategic choice.

The core tension here is transparency. Muse was marketed as an AI agent, yet users whose calls were handled during the test had no apparent disclosure that a human contractor might be on the other end, or listening in. That raises overlapping questions about informed consent, data privacy, and whether the product's public framing was accurate. The performance finding adds a practical wrinkle: if humans outperform the AI in the very task the product is built around, the timeline for genuine autonomy is an open question.

The Pro-AI camp has largely set the conduct controversy aside, with Miramar News pointing to the strong investor enthusiasm Muse generated around the Connect keynote. The implicit argument is that early-stage hybrid approaches are routine and that market confidence reflects the product's long-term trajectory. Critics have been sharper. Bess Hamilton summarized a recurring concern with the observation that new robot servants keep turning out to be underpaid humans. Janet Vertesi framed the product's purpose skeptically, and Unfitforpurpose raised a direct security warning against granting Muse computer access. An internal voice captured by reporting, identified as davidee, described the model as transferring wealth to the few at great cost to the many, while another internal account urged Meta not to launch without making the human-handoff a user-visible preference. The middle ground camp has not weighed in publicly yet; a story like this would typically draw arguments from that camp about the legitimacy of hybrid human-AI pipelines during development, provided disclosure standards are eventually met.

The clearest near-term signal to watch is what Meta says officially at or after Connect: whether it discloses the hybrid arrangement, commits to transparency standards for future tests, and addresses the conduct complaint involving the contractor who made racist remarks.