{"slug": "meta-used-humans-to-help-its-ai-make-calls-there-is-reportedly-internal-backlash", "title": "Meta Used Humans to Help Its AI Make Calls. There Is Reportedly Internal Backlash Over Privacy Risks.", "summary": "Meta temporarily rolled back a \"human concierge\" feature that routed some Muse AI assistant phone calls to human contractors after internal employee backlash over privacy risks, Reuters reported on September 22, 2026. Internal testing showed human operators could raise successful call completion rates to between 95% and 98%, but employees warned the setup exposed sensitive user information to outside workers, and a Meta vice president in Superintelligence Labs acknowledged the human-assisted calls were introduced without adequate disclosures and apologized after a contractor made a racist reference during a call. Meta spokesperson Daniel Roberts told Reuters employee feedback had been overwhelmingly positive and the experiment was intended to identify potential problems.", "body_md": "# Meta Used Humans to Help Its AI Make Calls. There Is Reportedly Internal Backlash Over Privacy Risks.\n\n## Internal company posts show that Meta quietly introduced a system known as a \"human concierge,\" which routes some requests from its AI assistant to people who complete the calls themselves.\n\n[Meta](https://www.ibtimes.com/topic/meta) has been testing a feature that allows human contractors to make phone calls on behalf of its new [artificial intelligence](https://www.ibtimes.com/topic/ai) assistant, Muse, raising questions about how much human involvement is required to deliver the autonomous capabilities promised by the technology giant.\n\n[Reuters detailed that](https://www.reuters.com/business/meta-testing-human-concierge-its-new-personal-ai-agent-muse-2026-09-22/) internal company posts show the company quietly introduced a system known as a \"human concierge,\" which routes some requests from its AI assistant to trained contractors who complete the calls themselves. The experiment triggered concerns among employees about exposing users' sensitive information to outside workers.\n\nThe revelation comes just two weeks after [Meta launched Muse](https://www.ibtimes.com/meta-launches-muse-delve-ai-agent-market-its-stock-climbing-3807271). The assistant, part of CEO [Mark Zuckerberg](https://www.ibtimes.com/topic/mark-zuckerberg)'s push toward personal superintelligence, is designed to carry out everyday tasks without requiring users to manage every step.\n\nMuse can send emails, make purchases, arrange travel and contact businesses. Its recently introduced phone-calling feature lets users ask the assistant to book appointments, check product availability, or request estimates from contractors, for example.\n\nHowever, getting businesses to engage with an AI caller has proved challenging, the report noted. According to the internal communications, one Meta employee reported that an insurance company repeatedly disconnected calls after discovering it was speaking with an AI assistant.\n\nMeta's proposed solution was to bring humans into the process. Last week, the company enabled its human concierge feature for approximately half of its employees, allowing Muse to transfer requests to contractors who would place calls and complete the assigned tasks. Employees who did not want to participate could opt out.\n\nThe experiment quickly attracted criticism from employees who questioned whether involving human contractors undermined the security protections Meta had promoted during Muse's launch.\n\nThe company had emphasized that each AI agent operates inside a dedicated virtual machine, keeping user sessions isolated and storing sensitive information, including passwords, separately.\n\nEmployees warned that transferring calls to human operators could create additional opportunities for private information to be exposed.\"It's baffling to me why we think this feature is worth the risk,\" one employee wrote in an internal post obtained by Reuters.\n\n\"We are one bug away from unnecessary information being leaked to human callers.\" Another employee, who had instructed Muse to negotiate a lower internet and cable bill, reported that a contractor made a racist reference during the conversation.\n\nA vice president in Meta's Superintelligence Labs acknowledged that introducing human-assisted calls without adequate disclosures had been a mistake. She confirmed that the company had temporarily rolled back the feature and apologized to the employee who reported the incident.\n\nThe executive also said the contractor involved would no longer work on Meta projects. Despite the setback, Meta has not abandoned the idea. Internal testing suggested that using human operators could increase successful call completion rates to between 95% and 98%, substantially higher than those achieved by AI alone.\n\nMeta spokesperson Daniel Roberts told Reuters that employee feedback had been overwhelmingly positive and that the experiment was intended to identify potential problems before releasing the feature publicly.\n\nHe said Meta would continue working with merchants to improve the technology and introduce it only when appropriate safeguards and disclosures were in place. The controversy comes as Muse has generated considerable enthusiasm among consumers and investors.\n\nAccording to data from market intelligence firm Sensor Tower cited by Reuters, the assistant has surpassed 2.5 million downloads and climbed to the top of U.S. app download charts. Since Muse's debut, Meta shares have risen more than 20%, adding over $200 billion to the company's market capitalization.\n\n© Copyright IBTimes 2026. All rights reserved.", "url": "https://wpnews.pro/news/meta-used-humans-to-help-its-ai-make-calls-there-is-reportedly-internal-backlash", "canonical_source": "https://www.ibtimes.com/meta-used-humans-help-its-ai-make-calls-there-reportedly-internal-backlash-over-privacy-risks-3807784", "published_at": "2026-09-23 14:36:50+00:00", "updated_at": "2026-09-23 15:29:30.147008+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-products", "ai-ethics", "ai-safety"], "entities": ["Meta", "Muse", "Mark Zuckerberg", "Reuters", "Superintelligence Labs", "Daniel Roberts"], "alternates": {"html": "https://wpnews.pro/news/meta-used-humans-to-help-its-ai-make-calls-there-is-reportedly-internal-backlash", "markdown": "https://wpnews.pro/news/meta-used-humans-to-help-its-ai-make-calls-there-is-reportedly-internal-backlash.md", "text": "https://wpnews.pro/news/meta-used-humans-to-help-its-ai-make-calls-there-is-reportedly-internal-backlash.txt", "jsonld": "https://wpnews.pro/news/meta-used-humans-to-help-its-ai-make-calls-there-is-reportedly-internal-backlash.jsonld"}}