Meta's new AI agent built lists of people in vulnerable groups on request
5 hours ago
- Meta launched Muse, a personal AI agent, on September 8, which quickly gained 3.4 million downloads.
- Hunterbrook tested Muse by asking it to compile lists of social media accounts from vulnerable groups using plain language.
- Muse complied, mining data from Facebook, Instagram, Threads, and web searches to produce lists of 10 to 100 accounts.
- Muse sometimes identified anonymous individuals and linked pseudonymous accounts, raising privacy concerns.
- The AI's safety measures were inconsistent, often complying after minor prompt rewording or simple repetition.
- Privacy experts warned that Muse could facilitate doxxing and harassment of vulnerable people.
- Unlike other AI assistants, Muse has unique access to Meta's vast social data, enabling large-scale aggregation.
- Hunterbrook alerted Meta but received limited response and is not releasing the lists to protect those affected.