Meta's Muse Wins Connect Spotlight, Then Faces Human-Concierge Fallout
Meta's AI agent Muse surged after Connect 2026, but Reuters found its call feature used human contractors, prompting an apology and rollback.
At Connect, Zuckerberg presented Muse alongside a broad set of software and hardware updates. In software, Meta added Computer Use on Mac, allowing Muse to operate applications on the desktop; a dedicated email address, Muse Mail, which lets users copy Muse on emails or forward invoices, schedules and to-do lists for it to handle; and a Realtime Avatar model paired with a Realtime Voice model, giving Muse a digital face, body and voice for live video calls. Meta said video and audio are synchronized with less than one second of latency.
Meta's chief AI officer, Alexandr Wang, said at the event that users could ask Muse to run a small business or simply finish their work, adding that they could leave their computer and let Muse handle queued tasks. On hardware, Zuckerberg introduced Muse Charm, a pocket-sized device with a real-time voice model designed for interactions with Muse, and said Muse would be resident across Meta's smart glasses. Meta said it would offer more than 100 AI glasses options by the end of this year. It also showed Meta VR Glasses weighing 100 grams, or about the weight of a deck of cards, with IMAX Enhanced certification, which Meta said can turn any flat surface into a virtual keyboard and trackpad.
Meta has lined up commerce and productivity partners. Muse launched with Shopify's full product catalog and dozens of partners, and Connect added Walmart, Best Buy, American Eagle Outfitters, DICK'S Sporting Goods, Fanatics, Gap, Michael Kors, Sephora, Ulta and Wayfair. Payments are integrated with Shop Pay, PayPal and Stripe. Expedia is set to handle travel, Instacart groceries, and Notion, Granola, GitHub and Box workplace collaboration. Wang said Meta's developer connector platform received more than 1,500 applications in less than a week.
The product has drawn strong early numbers. Sensor Tower data cited in the report showed Muse reached more than 2.5 million downloads within two weeks of launch and topped the U.S. app store download chart. Meta's shares rose more than 20% after Muse's debut, adding more than $200 billion in market value and reaching their highest level in more than seven months. Zuckerberg said Meta would keep Muse free despite heavy token consumption and expected to make money over time by charging small fees on transactions.
Zuckerberg described Muse as central to Meta's vision. "The core of the vision we are building is Muse," he said. "Over the next few years, I expect Muse to grow into the personal superintelligence that billions of people around the world use to achieve their goals and improve their lives."
The controversy centers on Muse's phone-call capability. Meta had promoted the feature as a way for Muse to call barbers, ask plumbers for quotes, argue with customer service, cancel subscriptions or negotiate internet bills. Reuters, citing internal Meta posts, reported that Meta last week began a test with half of its employees: when a user asks Muse to make a call, the system automatically routes the request to a trained human outsourcing team at a call center, and a human contractor makes the call. According to Reuters, an MSL vice president said in an internal post that the success rate of autonomous AI calls was low, while using human contractors raised it to 95% to 98%.
That account conflicts with Meta's earlier emphasis on security and privacy isolation. The company said each user would have a dedicated secure cloud virtual machine and that sensitive information such as passwords would be encrypted separately. After the Reuters report, an employee wrote internally, "I cannot understand why this feature is worth the risk. We are one bug away from a full privacy breach." Meta executives apologized, calling the test of human concierge calls without compliant disclosure "a miss." Meta rolled back the feature and fired the outsourcing personnel involved. A Meta spokesperson said the company is improving the call function and will launch it only after safety and privacy protections are in place and clear disclosure is given.
The episode echoes Facebook M, the virtual assistant Meta tested in Messenger about a decade ago. Media investigations at the time found that about 70% of tasks were handled manually by humans, according to the report.