Meta's Muse AI Agent Secretly Uses Human Contractors for Phone Calls, Sparking Privacy Concerns

Meta's newly launched Muse AI agent, positioned as an autonomous AI personal assistant, has been quietly relying on human contractors for AI calls—a revelation that has sparked privacy concerns and internal backlash among the company's own employees1

1

2

2

.

Human Concierge Feature Raises Privacy Questions

Meta informed employees last week about a human concierge test for Muse, which involves human agents behind AI handling phone calls placed through the digital agent1

1

. The feature was enabled for half of Meta's staff, with an opt-out option available3

3

. According to internal posts, "Muse is now able to hand requests to a trained agent, who places the call and works it through"3

3

.

The revelation has triggered immediate privacy concerns within Meta. Employees warned that sensitive information could be shared unintentionally with call center contractors, potentially exposing user data to third parties1

1

2

2

. One employee urged leadership to reconsider, stating: "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"5

5

.

Racist Remark Incident Highlights Ethical Hurdles

The human intervention in AI took a troubling turn when one Meta employee asked Muse to call an internet and cable provider to negotiate his bill. A transcript revealed that the human agent layer involved had made a racist reference during the call2

2

. Meta apologized to the employee and confirmed the contractor would not work on any of its projects again2

2

. A Meta Superintelligence Labs vice president acknowledged that introducing human operators without clear disclosure was "a miss" and announced the human concierge feature had been rolled back4

4

.

Why Meta Turned to Human Contractors for AI Calls

Meta's internal testing reveals the practical limitations of AI autonomy. Some businesses simply hang up on Muse when they realize they're talking to an AI agent2

2

. By handing AI agent phone calls to human agents, Meta found it could increase the success rate to between 95% and 98%, compared with a much lower rate for calls handled entirely by AI2

2

4

4

.

The phone-calling feature allows users to contact US businesses for tasks such as booking haircuts, checking whether an item is in stock, or obtaining quotes1

1

3

3

. Muse, launched on September 8, has reached 2.5 million downloads according to Sensor Tower data3

3

.

Employee Feedback and Negative Publicity Fears

Meta employees expressed deep concerns about potential negative publicity. "Not sure what I am missing here but why is this a 'feature?' 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 wrote on an internal message board5

5

. Another pleaded: "Please think about how the headlines will scream 'Meta uses humans behind the scenes to make calls' or 'Muse AI is actually independent contractors' and result in a ton of bad press about the privacy and security of our data handling"5

5

.

Meta spokesperson Daniel Roberts said employee feedback had been "overwhelmingly positive" and emphasized that internal testing was meant to "get feedback so we can implement safety and privacy protections and improve features before we release them publicly"1

1

4

4

. He added that Meta would "only roll it out when it's ready and with the proper disclosures"1

1

3

3

.

History Repeats: Facebook M's Failed Experiment

This isn't Meta's first attempt at blending human agents behind AI with digital assistance. The company previously launched Facebook M in 2015, a Messenger assistant supported by human trainers3

3

2

2

. Every message was reviewed by contractors at Meta's Menlo Park headquarters2

2

. Facebook M remained in limited beta and was shut down in 2018, never achieving more than about 30% automation3

3

.

Broader Pattern of AI-Driven Service Limitations

Meta joins a growing list of companies whose supposedly automated services rely heavily on human intervention. Amazon's "Just Walk Out" stores relied on more than 1,000 workers in India to review transactions and label footage, with around 700 out of every 1,000 purchases requiring human review as of 20222

2

. Presto Automation revealed that around 70% of orders handled by its AI drive-thru system required intervention from off-site human agents2

2

.

What This Means for AI Personal Assistant Development

The Muse situation exposes fundamental tensions in AI personal assistant development. While Meta emphasizes that each agent operates in its own virtual machine with a separate Sentinel agent checking outgoing actions2

2

, these technical safeguards say nothing about human contractors using headsets3

3

. The company claims conversations and user data stored in the virtual machine are kept out of its advertising systems2

2

, but the introduction of call center contractors adds new data privacy complexities.

Watch for how Meta addresses transparency around human involvement, whether competitors face similar AI-driven service limitations, and if regulatory scrutiny increases around disclosure requirements for AI services that rely on human intervention.