2026-09-23 / Signal #1
Meta’s Muse AI is secretly handing your phone calls to human contractors
“Your new AI assistant just put a stranger on the line and didn’t tell you.”
The Story
Meta tested a “human concierge” for its new Muse personal AI agent, quietly routing some user-requested phone calls (haircuts, bill negotiations, contractor quotes) to human call-center contractors instead of the AI. Employees flagged privacy leaks, a racist slip in one transcript, and the contradiction with Muse’s “secure VM” marketing; Meta rolled the test back for now after calling it “a miss,” while noting humans hit 95-98% success versus lower AI rates. Muse has topped US download charts since its Sept. 8 launch.
Why It Matters
The future of “autonomous” agents is still a call center with extra steps—hidden human labor, identity slippage, and the old Facebook M playbook repeating itself in 2026. Perfect collision of AI promises, work, privacy, and institutional sleight-of-hand.
Evidence
Reuters exclusive (internal Meta posts)
Sources
Daily scan: 2026-09-23