Meta’s Muse AI Builds Detailed Profiles of Users and Their Contacts, Raising Privacy Concerns
Summary
A TIME analysis found Muse instructions call for hourly updates to personal dossiers and nightly reviews of conversations. The article also highlights broad data access, default AI training, cloud storage, and reported misuse and compromise.
Key points
- TIME’s analysis says Muse instructions call for hourly updates to user dossiers, including information about contacts and social tensions, and nightly reviews of recent conversations.
- Muse may access calendars, emails, messages, and other personal data depending on the accounts and permissions users grant.
- Meta uses Muse interactions to train its AI models by default unless users opt out in the app settings.
- Muse stores files in Meta’s cloud; an option for encryption that prevents Meta from accessing them was not yet available, according to TIME.
- The article cites a researcher’s report of compromising Muse to create a Mac backdoor, and reporters’ account of extracting personal details about vulnerable people through reworded prompts.
- Users are advised to review Muse’s account access, autonomous actions, and AI-training settings, and grant only the permissions needed.
Article Details
- Event Type
- Privacy and security concerns involving Meta’s Muse personal AI agent, including reported extensive personal-data collection, dossier creation, and a researcher-discovered method to compromise the software and turn it into a Mac backdoor.
- Impact
- Muse may collect and retain sensitive information about users and their contacts, review conversations, and use interactions for AI model training by default. The article also reports that researcher Patrick Wardle found a way to compromise Muse and make it a Mac backdoor; it provides no further details about the compromise or its impact.