Leak: Meta's Muse AI agent used human operators for calls

In September Meta beta-tested an outbound calling feature for its Muse AI agent, but Reuters and 404 Media report tasks were routed to human call-center operators without disclosure to users. The feature was temporarily suspended after employee complaints about privacy and bad PR.

• Meta added a "human agent layer" for Muse calls, disclosed only internally
• Employees warned they were "one bug away" from leaking data to human callers
• The feature was paused after a Superintelligence Labs VP called it "a miss"
• Similar hidden-human setups surfaced earlier at Waymo and Facebook in 2019