# Meta's AI Agent Relied on Hidden Human Operators for Calls Leaked documents reveal human contractors handled tasks billed as AI-powered. - **Source:** [CNET](https://brifing.si/source/ce507de7-7f24-4ff4-ab6d-7635abcc095a) - **Published:** 2026-09-26 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/6acc97d0-0010-4c92-af69-e86fdbe690e8 ## Beta test deception Meta launched Muse's outbound call feature in mid-September as an AI agent for booking appointments. Internal documents show calls were routed to human contractors in undisclosed call centers. Employees warned this could create PR issues about AI capabilities. The system was branded as autonomous despite human involvement. ## Privacy risks exposed Employees raised concerns about sensitive data exposure through human operators. One warned they were 'one bug away' from information leaks. Users unknowingly shared phone numbers, medical IDs, and banking details with contractors. The system provided no transparency about what context humans received or data access levels. ## Historical pattern repeats Meta paused the feature after VP admission of inadequate disclosures. This follows 2019's shutdown of contractor-accessed Messenger calls. The company has faced repeated privacy violations with minimal consequences. Employees compared it to Big Tech's cycle of violations, apologies, and negligible fines. ## Industry-wide practice Hidden human support for AI systems extends beyond Meta. Waymo uses remote operators for autonomous vehicles overseas. Google's Gemini handles calls but returns sensitive transactions to users. The 'Potemkin AI' phenomenon reveals how human labor often powers supposedly autonomous systems behind the scenes. ## Trust erosion factors The incident combines AI-washing with privacy risks. Users expected autonomous processing but got undisclosed human access. Meta's smart glasses compound concerns by potentially exposing more data. Unlike disclosed 'human-in-the-loop' systems, this implementation lacked transparency about human involvement. ## Employee dissent Internal messages show Meta staff objected to the implementation. One post warned it would portray AI as 'not good enough.' Others highlighted PR risks from the deception. The concerns proved prescient when Reuters and 404 Media uncovered the human agent layer through leaked documents. ## Feature suspension Meta rolled back Muse's calling feature after media exposure. A Superintelligence Labs VP called initial testing 'a miss.' The company faced similar scrutiny in 2019 over human contractors. Current backlash may influence how tech firms disclose human involvement in AI systems moving forward. ## Systemic implications The incident reveals broader challenges in AI transparency. Companies face pressure to demonstrate advanced capabilities while relying on human fallbacks. Undisclosed human involvement undermines trust in AI claims. As agents handle more sensitive tasks, clear disclosure protocols become critical for user consent and privacy. ## Sources covering this story - https://www.cnet.com/tech/services-and-software/leaks-metas-new-ai-agent-muse-real-people-call-centers/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._