Every Mac comes with an (undocumented) family member classifier

Your Mac now secretly profiles your social network and daily activities for potential AI use.

Since macOS 13 (Ventura) in 2022, all Macs have included two undocumented daemons that build a personal knowledge graph about the user and their social connections. These daemons, knowledgeconstructiond and intelligenceplatformd, collect data from Contacts, Calendar, Photos, Messages, Mail, location history, Screen Time, and various behavioral signals. They construct a detailed graph stored in plaintext SQLite databases within ~/Library/IntelligencePlatform, accessible with Full Disk Access. This graph maps relationships between people, places, organizations, and events, storing inferred connections like family ties. Researchers discovered this system during security analysis of Apple Intelligence components and have released a tool called iKnow to allow users to inspect their own graphs. The data includes communication logs, location, and device signals, creating a rich profile of user activity and social interactions.

This discovery reveals that Apple has been building an extensive personal knowledge graph on user devices for years, long before the explicit announcement of Apple Intelligence features. The system's ability to infer complex relationships, including family ties, from disparate data sources like Contacts, Photos, and communication logs indicates a sophisticated on-device AI capability. The existence of this undocumented graph raises significant privacy implications, as it forms the foundation for future personalized AI experiences and could potentially be a single point of failure for sensitive personal information if compromised.

The open-sourcing of the iKnow tool is a crucial development, empowering users to audit the data collected about them and their social circle. This transparency is vital given the sensitive nature of the information stored and the potential for misuse. It also signals a broader trend of AI systems requiring comprehensive personal data for functionality, pushing the boundaries of on-device processing and data inference, and prompting questions about user consent and data control in the age of pervasive AI.