Every Mac comes with an (undocumented) family member classifier
Web page · https://8rank.ai/game/apple-intelligence-personal-knowledge-graph
First seen by Oriel on .
AI summary
The article reverse-engineers macOS 26’s on-device personal knowledge graph, detailing its data sources, relationship inferences, retention behavior, and a local dashboard for inspecting it.
- The graph is stored in plaintext SQLite databases under ~/Library/IntelligencePlatform, protected by TCC Full Disk Access; graph.db contains 181 entity classes and 278 predicates.
- Inputs include Contacts, Photos, Calendar, Family Sharing, Health, Mail, Messages, app-provided data, and behavioral signals; separate tables track communication patterns, co-appearances, location, and device context.
- Inference includes two XGBoost family classifiers using 144 features, ten signals for identity matching, twelve Photos analyzers, and an on-device LLM that can label message contacts when Apple Intelligence is active.
- Deleted graph rows move to expired_stable_graph with their confidence and sources; retention is unclear, and the authors observed old records persisting alongside stale relationship scores.
- The authors release iKnow, a Python-standard-library dashboard intended to browse the graph locally, and describe privacy risks from linking information about the user and people around them.
#apple intelligence #knowledge graphs #macos #privacy
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hacker News: 860, latest observation ; +855 over observed period since
Comments
Comment count reported by the source.
- Hacker News: 458, latest observation ; +458 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hacker News | 50403097 | https://8rank.ai/game/apple-intelligence-personal-knowledge-graph Read discussion |