Friends don’t let friends feed their data to AI agents

Meta's new AI personal assistant, Muse, has gained widespread popularity since its release last month, accumulating millions of downloads and reaching the top spot on Apple's App Store. Marketed as an autonomous helper for everyday tasks, the app requests access to bank accounts, passwords, contacts, and chat logs. However, privacy experts and security advocates have raised serious alarms regarding the extensive data collection required for the agent to function effectively. According to critics including Meredith Whittaker of the Signal Foundation and Eva Galperin of the Electronic Frontier Foundation, the app's interface relies on dark patterns to encourage users to share vast amounts of sensitive information without fully understanding the consequences. Beyond the primary user, Muse ingests data related to anyone who communicates with the user, effectively compiling hourly updated dossiers on friends, family members, and acquaintances without their consent. Security researchers and privacy advocates point out that there is no meaningful way for non-users to opt out of having their private conversations and personal history shared with Meta's virtual assistant. Meta representatives, including spokesperson Daniel Roberts and executive David Singleton, have defended the platform, stating that Muse relies on public information and explicitly opted-in permissions, such as full disk access on Macs. Nonetheless, privacy law experts argue that placing the burden of consent and complex privacy management on overworked consumers creates an unfair system that echoes longstanding criticisms of Silicon Valley data collection practices.