Apple's Hidden 'Family Classifier' Built Into Every Mac
Original: Every Mac comes with an (undocumented) family member classifier
Why This Matters
An undocumented, always-on personal graph feeding on-device AI raises real privacy accountability questions.
Since macOS 13 (2022), every Mac ships two undocumented daemons—knowledgeconstructiond and intelligenceplatformd—that build a queryable personal knowledge graph by cross-referencing Contacts, Mail, Photos, Messages, location, and behavioral data.
Security researchers at Eighth Rank have published a detailed breakdown of Apple's Personal Knowledge Graph, a system silently present on every Mac since macOS 13 Ventura. Two background daemons—knowledgeconstructiond and intelligenceplatformd—continuously analyze on-device data sources including Contacts, Calendar, Photos, Messages, Mail, location history, Screen Time, and additional behavioral signals to construct a structured graph of who you are, who you know, and how those people relate to each other.
The system's man page offers only a single-paragraph description, and Apple has published no public documentation on what exactly is collected or inferred. A 2023 Apple research paper outlined some underlying techniques, but said nothing about macOS integration specifics or the system's most striking capability: kinship inference—the ability to classify relationships between people in your graph (e.g., identifying family members) without explicit user input.
The graph functions as a queryable store that Apple Intelligence LLMs reportedly retrieve from at runtime. Researcher Nic Lydon first spotted the database earlier in 2026; Eighth Rank's analysis goes deeper, examining what the graph actually holds and demonstrating the attack surface it creates. They've released an open-source local dashboard called iKnow, letting users inspect their own graph's contents. The post also previews a threat-modeling section on how such a personalized, queryable store could be weaponized.