Meta tested having human contractors in call centers quietly handle a share of the calls users asked its Muse AI agent to make, according to internal posts seen by Reuters. Meta turned on these “human agent calls” last week for half its staff, after businesses kept hanging up on Muse once they realized an AI was calling. A Superintelligence Labs vice president said internal tests showed humans could complete 95% to 98% of calls, but she also conceded it had been “a miss” to begin testing without the right disclosures, adding that Meta had rolled the test back. Employees had warned that sensitive details could leak to contractors, and one found a contractor made a racist reference on his call. Meta said it will launch the feature only once it’s ready and properly disclosed.
