Meta’s Muse agentic platform launched on September 8, and has earned enough buzz to top the iOS download chart in the U.S. That’s an indicator of significant curiosity if nothing else—perhaps helped along by the announcement last week that a Muse beta feature can make phone calls on behalf of users.
we’re expanding our ☎️ 📞 beta for muse! https://t.co/ZJnXM2vb4U
— Alexandr Wang (@alexandr_wang) September 16, 2026
A Reuters report seems to confirm that these calls have been happening, but at the same time, that report is unlikely to bolster anyone’s faith in the power of AI agents. Some Muse tasks, Reuters says, have involved a “human concierge” feature that allowed a human to step in when the AI voice wasn’t up to the task.
Internal conversations at Meta apparently seen by Reuters seem to shed some light on why this may have been deemed necessary: one user apparently said when he tried to get the Muse bot to call his insurance company, the company repeatedly hung up when the bot disclosed that it was not human. When the human calling feature was added for Meta’s employee testers, they were apparently allowed to opt out.
But further internal conversations apparently also questioned the wisdom of such a feature, which some employees are apparently worried could feed sensitive user information to a human operator. Reuters describes the feature as able to “handle errands like booking haircuts, checking whether a store has an item in stock or getting quotes from different contractors.” It’s not necessarily invasive if a human contractor calls your barber, but it could be invasive if they called—and this is me coming up with my own example here—your urologist.
What’s more, one of the employees who apparently tested the human calling feature reviewed a transcript of a call, and claimed that the human operator involved “had made a racist reference,” to use Reuters’ phrasing.
Amid these internal communications, a Meta Superintelligence Labs vice president apparently apologized to the employee who complained about a contractor allegedly getting racist on a phone call, and explained that the company was no longer working with that individual. They said, per Reuters, that the company had “rolled back this feature,” noting that the decision to test calls from human contractors the way it had was “a miss.”
Meta’s tests, the VP apparently claims, show that human involvement could drive the phone call success rate up into the high 90% range instead of the “lower percentage of AI calling.”
Gizmodo sought clarity from Meta about this “lower percentage of AI calling” quote. It’s not completely clear from Reuters’ report whether all Muse-powered calls are rolled back, or just—as seems more likely—the ones involving human contractors. If calls are still happening, but none involving human contractors, it’s not clear from Reuters what, if anything, is happening to compensate for AI calls being rejected.
In response, Meta spokesperson Dave Arnold gave us the following statement, saying it was originally provided to Reuters: “Internal testing, aka dogfooding, is core to the product development process. While the response from employees has been overwhelmingly positive, the entire point is to get feedback so we can implement safety and privacy protections and improve features before we release them publicly. We’re working with merchants to continue improving this potential calling feature, and will only roll it out when it’s ready and with the proper disclosures.”
Read the full article here
