Meta has rolled back, for now, an internal test that let human contractors handle some calls requested through Muse, its new personal AI agent, after employees raised privacy and disclosure concerns, according to Reuters.
Meta launched Muse on Sept. 8 as a personal AI agent designed to handle tasks such as sending emails, shopping, booking travel and, in Meta’s launch language, turning “long-term goals into action plans.” Meta also emphasized that Muse runs inside a dedicated virtual machine in its cloud and that users control which services the agent can access.
Days later, Meta expanded a beta for outbound calls to U.S. businesses. TechCrunch reported on Sept. 17 that Muse had gained the ability to place calls, a feature positioned alongside similar call-handling tools from rival AI agents.
Reuters reported that Meta then tested a “human concierge” system in which contractors could quietly handle some calls placed through the digital agent. The phone feature was meant to handle errands such as booking haircuts, checking whether a store has an item in stock or getting quotes from contractors.
According to Reuters, the human concierge option was enabled for half of Meta employees testing Muse last week. Employees who did not want to be included could join an opt-out group, the report said.
The privacy issue
The concern was not only whether Muse could complete calls. It was whether sensitive instructions intended for an AI agent could be passed to a human contractor without clear notice before the call happened.
Reuters said employees warned that call requests could expose sensitive information to contractors in call centers. One employee reportedly wrote that Meta was “one bug away from unnecessary information being leaked to human callers.”
Another employee who asked Muse to call an internet and cable provider to negotiate a bill said a transcript showed the contractor made a racist reference during the call, according to Reuters. A Meta vice president later acknowledged internally that it was “a miss” to test contractor-placed calls without proper disclosures and said Meta had “rolled back this feature” for now.
Meta says broader release depends on disclosures
Meta spokesperson Daniel Roberts told Reuters that employee feedback had been “overwhelmingly positive” and that the purpose of the test was to gather feedback so Meta could add safety and privacy protections before public release.
Roberts said Meta was working with merchants to improve the potential calling feature and would roll it out only when it was ready and had proper disclosures.
404 Media, which also reported on Meta internal posts, quoted a Meta spokesperson saying internal testing is core to product development and that feedback is used before features are publicly released.
Why the distinction matters
AI agents are being pitched as software that can act on behalf of a person, not merely answer a question. That shift raises the stakes for consent, audit trails, data access and responsibility when a task leaves the chat window and reaches a real business by phone.
For a simple stock check, human help may seem low-risk. For calls involving bills, healthcare, legal matters, employment, financial accounts or family details, the difference between an automated agent and a contractor in a call center can matter.
The issue also cuts in the other direction. Businesses receiving calls from AI agents or contractors acting through AI systems need to know who, or what, is on the line so they can decide how to verify requests, handle consent and record notes accurately.
A familiar problem in AI launches
Reuters compared the Muse test with Facebook’s earlier M assistant in Messenger, which relied heavily on human operators during its experimental run. The comparison shows how autonomous AI demos can depend on hidden human work while companies test whether the technology can be made reliable enough for everyday use.
The business reason is clear: a phone call that gets completed feels more useful than one that fails. Reuters reported that an internal Meta post said some tests indicated human callers could bring success rates into the 95% to 98% range, compared with a lower rate for AI calling.
That leaves the disclosure question unresolved. If a product is presented as an AI agent, users, contractors and businesses may all need clearer signals before sensitive details are shared.
What comes next
The immediate question is whether Meta brings the human concierge idea back with opt-in controls, clearer pre-call notices, stricter limits on what contractors can see, or some combination of all three. Meta has not publicly said the contractor-assisted version will launch.
For now, the Muse episode shows how quickly AI agent features can move from software capability to privacy, disclosure and labor questions. The feature may be useful if it saves time and completes frustrating calls. It also needs disclosure clear enough that users and businesses understand who is actually speaking before the conversation starts.
Get new small business insights by email
Practical ideas and useful articles to help you make better business decisions.
Tech Help Canada Staff researches, writes, and reviews practical content for business owners and professionals. Our coverage spans business, marketing, SEO, technology, and the tools and systems people use to grow and operate online. We focus on clear, useful information backed by research, hands-on experience, and editorial review. Learn more about our team and editorial standards. Need help with something? Contact Us
