How Apple's Personal Knowledge Graph learns about your life on Mac

Apple's Personal Knowledge Graph is a system used by the company to compile and analyze various data points about its users and their lives, as detailed in a breakdown by security researchers. This system has been in use since macOS 13 in 2022, with the generation process involving two daemons that analyze device content and create a knowledge graph. The daemons store information in a protected area on the Mac, within the Library/IntelligencePlatform directory, which contains 51 SQLite databases.

One of the key databases, graph.db, contains 181 entity classes and 278 predicates, with each row representing a subject, an object, and a relationship between the two. Subjects can include entities such as people, groups of humans, or organisms, while objects are typically plain values like names or dates. Predicates describe the nature of the relationship between the subject and object, such as what kind of thing it is, its genre, or topics of interest.

The daemons continuously populate this database with relationships over time, using various data points from sources such as messages, emails, and photos. For instance, it can infer relationships between people based on photos, communication logs, and physical proximity. While the researchers did not find the ContextualUnderstanding.PhysicalSocial data filled in, it appears to be a list of people and contact details that the user has been physically near at some point.

The daemons also handle deleted data by retaining it in an expired_stable_graph table, which contains retracted rows with confidence and relationship scores. This data can still outlive the information that helped produce it, as a conclusion derived from the models may persist even if the original data is deleted. Overall, this system raises privacy concerns due to the amount and granularity of data stored locally on the Mac, but it is ultimately more private than similar AI-centric services that analyze data in the cloud.