Meta used humans behind Muse calls: your business lesson
Direct answer: Meta tested having human contractors quietly make Muse's phone calls. Effectiveness rose to 95–98%, but employees flagged privacy risks and one incident involved a racist remark. Meta pulled the feature. The lesson: automation that promises full autonomy while hiding humans behind the curtain breaks trust. The model that works is AI plus declared human supervision.
Two weeks ago Meta launched Muse, its personal AI agent, with a star feature: making phone calls for you. You ask it to call the salon, check stock at a store, request quotes from contractors. The agent calls, talks, and gets it done.
The problem: last week it came out — via internal posts obtained by Reuters — that Meta was testing a "human assistant." Meaning: human contractors quietly making some of those calls on behalf of the digital agent.
Why they did it
Because it works. Internal tests showed that with real people making the calls, the success rate climbed to 95–98%, vs. the lower rate of AI-made calls.
And AI alone was hitting reality: one employee shared that he tried to get Muse to call his insurance company and "they hang up constantly when they hear Muse is an AI." People still prefer talking to humans on the phone.
Nothing new under the sun, by the way. A decade ago Meta — then Facebook — tested an assistant called "M" inside Messenger. It later came out that people handled around 70% of tasks.
Why it blew up
Employees raised the alarm, and rightly so:
- Privacy: real people listening to calls with users' sensitive information, with the risk of it leaking to call-center contractors. "We are one mistake away from unnecessary information being leaked to human operators," one employee wrote.
- One serious incident: another employee asked Muse to call his internet provider to negotiate the bill. The transcript showed the human contractor made a racist remark during the call.
- It contradicted their own promise: Meta had sold Muse on security — each agent with its own secure "virtual machine," sensitive data isolated. The human assistant undercut its main selling point.
A Superintelligence Labs VP admitted it "was a mistake" to start the test without proper disclosures and said the feature was pulled for now. She apologized to the affected employee and promised that "human agent" would never work on any Meta project again.
The official spokesperson, Daniel Roberts, said employee response had been "overwhelmingly positive" and that the goal was "to get feedback to implement safety and privacy measures before public launch." Translation: they tested it in secret, got caught, and are now dressing it up as process.
The lesson your business should actually take
Here's what almost nobody says out loud: the hybrid model is the right call. The secrecy was the failure.
An AI needing human supervision is simply the reality of the technology today. The 95–98% effectiveness proves it: human + AI wins. Meta's mistake was selling total autonomy while hiding people behind the curtain.
And that's exactly the trap many businesses fall into when they automate:
- They promise "100% automated" and fall short. The client finds out, and trust breaks harder than if they'd been honest from day one.
- They put customer data in hands they don't control. If your automation touches sensitive information, you need to know exactly who — or what — touches it.
- They mistake automating for disappearing. Good automation keeps the human where they create the most value, and lets AI handle the rest.
At inicIA we work exactly this way: AI agents that do the heavy lifting, with declared human supervision where it matters. No "total autonomy" theater. Because customer trust is worth more than an impressive demo.
Frequently asked questions
What is Muse's "human assistant"?
A Meta test in which human contractors quietly made the phone calls users asked Muse, the AI agent, to make. The feature was pulled after internal privacy complaints.
Why did Meta use humans instead of AI for the calls?
Because effectiveness rose to 95–98% with humans, vs. a lower rate with AI. Plus, many businesses hung up when they detected the caller was an AI.
Does Muse still make calls?
Meta said it keeps "improving the AI calling feature" and will only roll it out "when it's ready and with proper disclosures."
What should my business learn from this case?
That honest automation (AI + declared human) builds trust and delivers results; automation that hides humans behind promises of total autonomy creates crises.
Want to automate your business without these mistakes?
At inicIA we design agents and systems with human supervision where it matters — no false promises. Message us on WhatsApp and let's talk.