Meta's Muse AI Agent Wins Users but Demands Total Data Access
Meta's new agentic AI platform Muse has surged to 6.6 million downloads and 1.8 million daily active users less than a month after launch, but its aggressive data collection practices raise significant privacy concerns.
Meta has vaulted into the front rank of consumer artificial intelligence with Muse, an agentic AI platform that has amassed 6.6 million downloads and 1.8 million daily active users less than a month after its September 8 launch. The rapid adoption marks a striking reversal for a company that many observers had written off as a laggard in the AI race, and analysts at Citi estimate the platform could generate as much as $27 billion in revenue in the near term.
Unlike conventional chatbots that function largely as glorified search engines, Muse is designed to take concrete actions on a user's behalf. It can write and send emails, book travel, fill out forms, and sign into websites. Meta markets the service with the tagline «AI that gets it done» and encourages users to hand it tasks such as booking a flight or registering for a road race. The technical foundation for this capability is a virtual machine in the cloud assigned to each user, an approach pioneered by the techie-oriented agentic platform OpenClaw. That virtual computer allows Muse to browse complex websites, log into accounts, and operate independently even when the user is not actively chatting with it.
The platform's personality is as distinctive as its capabilities. Muse is personified by a fluffy, smiling mascot that appears on billboards and at the top of the app during conversations. Users can customize their Muse avatar and give it a name, though «Mark Zuckerberg» is reportedly not permitted. The bot communicates in a friendly, casual style full of first-person pronouns and text-message conventions, a deliberate contrast to the more formal tone of competitors like ChatGPT. Experts have long cautioned against anthropomorphizing AI, but Meta has embraced the approach fully.
In practice, the agentic design produces results that go well beyond informational responses. In one test, Muse gathered lease quotes for a Kia when asked what car the user should consider. In another, it scanned financial data, determined that a Hulu subscription was overpriced, found better plans matched to a family's actual usage, and offered to switch the account, saving $80 per month. The bot also sends unprompted notifications, such as a report on how to lower an electric bill or advice on selling a used Tesla for top dollar.
That usefulness comes with an unusually aggressive appetite for personal information. From the moment the app opens, Muse requests full access to Gmail, permission to read text messages, and access to the photos on a user's phone. It seeks continuous location tracking through the device's GPS and encourages users to share conversations they have had with competing chatbots so it can «get to know» them better. The platform also presumably draws on the vast troves of data Meta has collected over decades of tracking user behavior across the web and its own services.
The privacy implications are substantial. Meta's history of extensive data collection means Muse may have access not only to what users explicitly share but also to detailed records of their online activity, including browsing habits and shopping cart contents. For users attracted by the platform's can-do attitude, the trade-off is clear: Muse is powerful, capable, and genuinely useful, but it works best when it is told essentially everything about a person's life. Whether that bargain proves acceptable to mainstream consumers will shape not only Meta's AI ambitions but also the broader debate over how much access agentic systems should be granted to the most sensitive corners of digital life.
4
