Meta Muse is People

Meta has been testing a "human concierge" for its new personal AI assistant Muse in which human contractors quietly handle some phone calls placed through the digital agent, according to internal company posts seen by Reuters. Meta told employees about the test last week, shortly after publicly launching a phone-calling feature for Muse, and some employees raised privacy concerns that sensitive information could be shared unintentionally with call-center contractors, with one writing, "We are one bug away from unnecessary information being leaked to human callers." Muse, which has topped US app download charts in the two weeks since its debut, is designed to autonomously send emails, shop and book travel, and Meta appears to have rolled back the human-concierge test for now.

Meta Muse is People Basically impossible for me to ignore this headline after yesterday's " Soylent Googlebook is Android https://spyglass.org/googlebook-android/ ". As Katie Paul reports: Meta has been testing a “human concierge” for its new personal AI assistant, Muse, which entails having human contractors quietly handle some of the phone calls placed via the digital agent, according to internal company posts seen by Reuters. The Facebook and Instagram owner told employees about the test last week, shortly after the company publicly launched a phone-calling feature for Muse. The Muse agent, which has topped US app download charts in the two weeks since its debut, is designed to autonomously perform tasks like sending emails, shopping and booking travel on a person’s behalf. With the phone-calling feature, a Muse user can tell the agent to dial phone numbers for US businesses and talk with people on the other end to handle errands like booking haircuts, checking whether a store has an item in stock or getting quotes from different contractors. On one hand, this could not be any less surprising. This type of outsourcing-AI-to-humans happened a lot in the previous iterations of AI. <sup 1</sup one As the story goes on to note, Meta's long-forgotten first attempt at a virtual assistant from over a decade ago <https://en.wikipedia.org/wiki/M virtual assistant ?ref=spyglass.org – so long ago that they were still called Facebook – 'M' was leveraging this hybrid system. <sup 2</sup two The difference, I suppose, is that everyone actually believes in the AI to not only work , but to be able to do these tasks autonomously now. It's quite the dichotomy to read headlines about AI being on the verge of destroying humanity https://spyglass.org/ai-end-of-humanity/ , while at the same time reading stories that it can't figure out how to make a phone call. The bigger problem here for Meta, of course, is the trust issue. Muse is a great product https://spyglass.org/metas-amusing-muse/ . That doesn't mean it will ultimately work as the 'AI personal assistant' holy grail product – a space with just as many corpses https://spyglass.org/the-swiss-assistant/ as any other over decades of attempts – but I think it has the best shot of anything I've seen to date. And again, that's in part because the AI is real this time. Well, real enough But, as my inbox can attest, a lot of people https://spyglass.org/why-didnt-google-build-muse/ have trouble envisioning using such a product from Meta – because it's from Meta . And so when there are headlines that actually the AI, which they promise has your data and information sandboxed away and safe, may be leveraging humans – contractors at call centers, no less – that feels... not great. Sort of like a bait-and-switch.<sup 3</sup And, well, Meta's own employees recognize this issue: Some employees have raised privacy concerns about humans handling those calls, warning that the approach could result in sensitive information being shared unintentionally with contractors in call centers, the internal posts showed. “It’s baffling to me why we think this feature is worth the risk,” one employee wrote in an internal post. “We are one bug away from unnecessary information being leaked to human callers.” To be fair to Meta here, this feature was very much in beta testing mode and not live for all Muse users. But they were also clearly rushing to get it out there, undoubtedly to compete with other offerings from products like Instinct https://techcrunch.com/2026/09/17/rival-ai-agents-instinct-and-metas-muse-both-add-the-ability-to-make-calls/?ref=spyglass.org which leads one to naturally wonder how they were pulling off the phone calling features... . Would they have disclosed the human-in-the-literal-loop if it launched? I mean, hopefully Instead, they decided to roll-it-back for now, it seems. The bigger picture remains that we're probably going to see more such hybrid approaches at the edges of AI as agents rise and try to crack real world tasks. The AI is better now, but there are still too many things in the real world that sort of require humans in some capacity – at least until the actual robots come. <sup 1</sup How about 'outsourcing-AI-to-I'? ↩ <sup 2</sup 'M' of course, reminds me of James Bond <https://en.wikipedia.org/wiki/M James Bond ?ref=spyglass.org , which seems fitting given my own Muse bot's name https://spyglass.org/metas-amusing-muse/ ... ↩ <sup 3</sup How about 'bAIt-and-switch'? ↩