Every Mac since 2022 runs two undocumented daemons that build a personal knowledge graph from your Contacts, Calendar, Photos, Messages, Mail, location, and Screen Time. This graph, stored in ~/Library/Intelligence Platform, is a queryable triple store with 181 entity classes and 278 predicates. It maps relationships, employment, and kinship, allegedly retrievable by LLMs. Despite Apple’s 2023 research paper, there is no public documentation on its integration, data collection, or inference methods.
Researchers Nic Lydon and others discovered it independently in 2026, revealing a privacy blind spot. The system stores data in plaintext SQLite databases, protected only by TCC Full Disk Access. A new open-source tool, 'i Know', lets users inspect their own graph.
The authors warn that if this knowledge base is compromised—especially when paired with an abliterated LLM—it could become a privacy.
Source: Hacker News · Summarized by HeadlinesBriefing