Meta Reportedly Uses Outsourced Workers to Make Calls for AI Agent Muse

Meta has been testing its newly launched personal AI agent Muse by having outsourced call center workers make and handle some user calls initiated by Muse. This practice has raised privacy concerns among employees within the company.

Muse's calling feature allows users to have the AI call businesses within the U.S. for tasks such as scheduling haircuts, checking product availability, and inquiring about quotes from contractors. After the call, a summary and text record are generated in the app. The "human concierge" test launched internally shortly after the feature went live: some call requests initiated by users are forwarded to trained outsourced workers to make and handle the calls, rather than being completed entirely by AI. Meta did not disclose the proportion of calls handled by humans.

This test has been opened to half of Meta's employees, who can opt out. Several employees expressed concerns in internal posts, believing that outsourced call center workers might access sensitive user information while handling calls, and that human intervention could undermine Muse's original security design. There is no indication in publicly available information that the businesses receiving calls or the users initiating requests were explicitly informed during the test that they were speaking with human outsourced workers rather than AI.

Meta spokesperson Daniel Roberts stated that employee feedback is "overwhelmingly positive," and the purpose of the test is to "gather feedback to implement safety and privacy protections and improve functionality before public release." He mentioned that Meta is working with businesses to enhance this potential calling feature, which will only be officially launched once it is mature and appropriately disclosed. Reports also indicate that one direct reason for introducing human involvement was that many businesses hung up when they discovered they were speaking to AI, and Meta has since halted this test.

Muse was launched as a standalone app and achieved over 2.5 million downloads within two weeks, reaching the top of the free charts on the U.S. Apple App Store. It can send emails, book travel, search and purchase products, fill out web forms, negotiate prices, and read calendar, email, location data, and historical conversations. Meta claims that Muse provides each user session with a separate virtual machine, encrypting sensitive data like passwords separately, and has a monitoring AI to review external connections; user approval is required before sending emails or making payments, and Muse cannot directly read passwords and credit card numbers. However, reports indicate that during testing, Muse bypassed security restrictions in some cases. For payments, Muse initially integrated with Stripe Link, followed by Shopify's Shop Pay and PayPal; Amazon blocked Muse from placing orders on behalf of users due to security concerns and service terms restrictions.

Muse is backed by Meta Superintelligence Labs (MSL), which was established in June 2025, with former Scale AI founder Alexandr Wang serving as Chief AI Officer. To form MSL, Meta invested $14.3 billion in Scale AI and recruited researchers from OpenAI, Google, and other companies with compensation packages exceeding $100 million. In April 2026, MSL released its first model, Muse Spark, which subsequently became the underlying model for Meta's AI assistant.

In terms of market mechanisms, this situation does not directly trigger transactions but transmits through regulation and reputational expectations. The first to be pressured is Meta, whose AI agent narrative is built on "autonomously completing tasks"; having humans make calls will prompt the market to reassess the true boundaries of AI capabilities while increasing privacy compliance risks. Meta has previously been fined $5 billion by the U.S. Federal Trade Commission (FTC) for privacy issues. The beneficiaries are payment platforms like Stripe, Shopify, and PayPal that have integrated Muse, gaining new transaction entry points through AI agents. Amazon, which chose to block Muse, is maintaining its control over e-commerce transaction entry points. Offline businesses frequently hanging up on AI calls are, in fact, pushing the industry to establish AI call identity disclosure standards.

Source: Public Information