Meta's Muse Draws Privacy Criticism After 900,000 Downloads in 1 Week
Updated
Updated · WIRED · Sep 20
Meta's Muse Draws Privacy Criticism After 900,000 Downloads in 1 Week
3 articles · Updated · WIRED · Sep 20
Summary
900,000 first-week downloads made Muse an early hit, but testing found the Meta assistant repeatedly pushed users to connect email, bank accounts and documents while lacking an option to fully disable its memory feature.
Meta says Muse gives users control and can complete tasks through a virtual machine that browses and clicks on the web, with stronger performance on shopping and Facebook Marketplace searches than many earlier AI agents.
By default, Muse also opts users into having interactions used for AI model training unless they switch off “Help improve our AI models,” a setup privacy advocates called a red flag despite Meta saying the data is sanitized.
Meta plans confidential virtual machines later this year to block its own access to in-session data, but critics warn the agent could still shape ads indirectly, steer purchases and deepen user dependence as it learns more personal preferences.