Meta's New AI Assistant Collects More Data Than It Delivers
Meta's Muse app, an AI assistant for everyday tasks, has been downloaded over 900,000 times in its first week. Testing revealed that the app aggressively prompts users to link bank accounts, email, and passport details, while its actual task performance is secondary to data collection. Meta maintains that user controls and protections are built in, but the author's experience suggests otherwise.
The app's rapid adoption—over 900,000 downloads in a week—is driven by Meta's cross-promotion on Instagram and WhatsApp. Its web-browsing "virtual machine" outperforms earlier agents, successfully navigating a bakery order and locating couches on Facebook Marketplace. However, the author found the experience less about task completion and more about data extraction.
Muse aggressively requests access to bank accounts, email, and passport details during setup. It maintains a "Memory" file of user preferences, though users can edit or request deletion. Meta insists protections exist, but the author's testing suggests data collection is the primary function, with task performance secondary. The "no tip" selection on a food order also highlights a lack of social nuance.
The widespread adoption of Muse could normalize the expectation that everyday AI assistants require deep access to financial and identity data. Users may trade convenience for surveillance, potentially eroding trust in AI if the experience feels extractive. This could particularly affect less tech-savvy individuals who may not grasp the implications of linking bank accounts. It may also prompt regulators to scrutinize how AI agents handle sensitive personal information, shaping future privacy standards.