Meta's Muse AI Calls Run Through Human Call Centers

Overview

Meta is expanding its Muse AI agent to include outbound calling capabilities, allowing users to have the assistant make phone calls on their behalf for tasks such as booking restaurant reservations or scheduling haircuts. While the company has publicly framed this as an advanced AI feature, internal documents reveal that the system relies heavily on human intervention. Meta is currently testing a setup where calls are placed by trained human agents in call centers, who act as a backend layer to complete requests that the AI cannot handle autonomously.

This revelation has sparked significant concern among Meta employees regarding privacy implications, data security, and potential public backlash. Internal communications show that staff members fear the disconnect between the product’s AI marketing and its actual human-powered infrastructure could damage user trust and invite negative press coverage. The move mirrors a broader industry pattern where emerging AI products rely on virtual workers rather than fully autonomous algorithms.

The Human Behind the AI

According to internal postings seen by reporters, the Muse calling feature operates through a hybrid model. When a user initiates a call request, the AI does not always manage the conversation end-to-end. Instead, it can hand the request off to a trained human agent who places the call and navigates the interaction with the business being contacted. The AI subsequently reports back with a transcript and summary of the completed task.

“Muse doesn’t just dial a number. It calls a business on your behalf, handles the conversation, completes your request, and reports back with a transcript and a summary,” an internal message board post explained. “It also no longer works alone. Muse is now able to hand requests to a trained human agent, who places the call and works it through.”

The feature was announced by senior executives, including Ryan Fox, principal engineer on Muse, and Alexandr Wang, Meta’s chief AI officer, on social media platforms in mid-September. They highlighted the expansion of the phone beta for US businesses. However, one employee noted that they were only informed a human was making the call after the interaction had already taken place during testing. This lack of transparency has raised questions about whether users are adequately informed when their personal requests are handled by contractors rather than software.

Privacy and Security Concerns

The involvement of human agents in processing user data has triggered alarm within the company. Employees have expressed worry that sensitive information—such as details about medical appointments or private scheduling—could be exposed to third-party contractors. One staff member argued that the company’s claim that these agents underwent extensive training was insufficient to guarantee data safety.

“That training is not a [security] mechanism,” one employee commented regarding the safeguards provided to human agents handling private user data.

Critics within the organization pointed out that users might assume their interactions with Muse are entirely private and automated. If a user asks Muse to schedule a confidential doctor’s appointment, they may not realize that a human contractor is making the call and potentially hearing the discussion. This scenario evokes historical precedents where tech giants launch AI products under the guise of automation, only for it to be revealed that humans are performing the bulk of the labor.

Internal debate also focused on the ethics of consent. Employees urged leadership to treat the human-agent feature as optional or to implement it as a user preference rather than a default setting. There was a strong sentiment that deploying this functionality without clear disclosure would erode the goodwill generated by early adopters.

“This is absolutely going to create a ton of outcry if we publicly launch this as default-on. It will kill all the goodwill and organic press we’re getting from early adopters,” one employee wrote in an internal forum post.

Public Reaction and PR Risks

While internal dissent is significant, public reactions to Muse’s calling feature have been mixed. Some early testers have praised the technology’s ability to navigate complex phone trees, wait on hold, and resolve issues autonomously. For instance, an AI researcher at Meta reported that when Muse called customer service, the representative did not seem to notice they were speaking to a bot, describing the interaction as “completely natural.”

However, other users have experienced failures, with some business representatives hanging up on the AI agent outright. Despite these technical growing pains, the more pressing concern for Meta appears to be the narrative surrounding the human element of the service. Employees fear that media coverage will focus on the fact that the AI is not sophisticated enough to operate independently, framing it as a cost-cutting measure disguised as innovation.

“This has potential for so much negative PR. It could portray us as ‘their AI is not good enough so they still need humans’ kind of coverage for this launch,” one employee warned.

The company has asked employees to keep details about the feature confidential until an official launch. Internal communications explicitly stated that the product is “confidential, pre-launch” and prohibited sharing any outputs or descriptions with outsiders. As Meta prepares for a broader rollout, the tension between the promise of autonomous AI and the reality of human-supported operations remains a critical challenge.