TL;DR
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
A report by Eighth Rank says macOS 26 builds an on-device knowledge graph from sources such as Contacts, Photos, Calendar, and communications metadata. Its analysis describes a classifier for family relationships, but Apple’s public documentation does not explain that feature or fully specify what data it uses.
Security research group Eighth Rank says macOS 26 builds an on-device personal knowledge graph that combines information from apps and system signals, including a classifier for family relationships that the group says is not documented by Apple. The report describes how the graph stores links between people, places, events and other entities; Apple’s brief public description says the system analyzes device content to build a general-purpose graph, but does not detail the reported kinship inference.
Eighth Rank says its analysis examined a Mac running macOS 26.6.2. It found two system processes, knowledgeconstructiond and intelligenceplatformd, working on the graph. Apple’s man page describes knowledgeconstructiond as analyzing on-device content to help build a graph representing the device and its user, with intelligenceplatformd handling some processing. The report says the graph and related data are stored in SQLite databases under ~/Library/IntelligencePlatform.
The researchers describe graph.db as a triple store: records connect a subject to an object through a predicate, such as a relationship between a person and an employer. They say each edge includes confidence, source and timestamp information. The report lists potential inputs including Contacts, Photos, Calendar, Family Sharing, Health emergency contacts, and suggestions drawn from Mail and Messages. Separate tables reportedly hold communication metadata, location and movement records, and other device signals. Eighth Rank says the communication log records who communicated, when, through which app and in which direction, but not message content.
The report’s headline refers to a family-member classifier, but the supplied material does not explain its precise operation, accuracy or full set of inferred relationships. Eighth Rank says its examination found little public documentation about how the graph is integrated into macOS or exactly what is collected and inferred. It also points to Apple researchers’ 2023 paper on graph-construction techniques, while saying that paper does not explain the macOS implementation or the kinship inference it describes.
How Relationship Inference Affects Privacy
A graph that links a user to people, events and places can make personal information easier for software to retrieve than a collection of separate records. If the report’s account is accurate, inferred family ties could expose relationships a user never entered as a single fact, while communication and location metadata could add further context. The sensitivity comes not only from individual records, but from how they are connected.
Eighth Rank warns that a consolidated personal graph could become a privacy single point of failure if accessed by someone unauthorized. That is the researchers’ risk assessment, not evidence in the supplied material that the databases have been breached or that the graph is routinely sent to a language model. The report says language models may retrieve from the store, but qualifies that claim as alleged. Whether and when that happens is not established here.
Products Worth Considering
Apple MacBook Air 13.6-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Sky Blue
Apple iPad 11-inch with A16 Chip, 512GB, Wi-Fi 6, and Apple Pencil (USB-C) — Blue
Kindle Paperwhite Kids (2024), 16 GB, 7-inch Display, Cyber City Cover
As an affiliate, we earn on qualifying purchases.
What Apple Documents About the Graph
Apple’s man page for knowledgeconstructiond provides a short description: the daemon analyzes content on a device to help build a general-purpose knowledge graph representing the device and its user, and works with intelligenceplatformd. That confirms the stated purpose of the process, but does not describe individual data sources, kinship classification or the graph’s complete contents.
Eighth Rank says Apple researchers published work on graph-construction techniques in 2023, but the report distinguishes that research from documentation of the feature as integrated into macOS. The group says it investigated the system on macOS 26 and released a local dashboard called iKnow to let users inspect graph contents on their own Macs. A separate earlier account by Nic Lydon, cited by Eighth Rank, also described finding a personal knowledge graph on a Mac.
“A daemon that analyzes content on the device to build a general purpose knowledge graph.”
— Apple’s knowledgeconstructiond man page, quoted by Eighth Rank
The Classifier’s Scope Remains Unknown
The supplied report material does not show the classifier’s rules, training or inference method, accuracy, or the exact family relationships it can identify. It also does not establish whether every Mac since 2022 runs the same components or collects the same signals; that broad claim appears in the report’s framing, while the detailed analysis described here was conducted on macOS 26.6.2.
Other open questions include how long each category of information is retained, which users or processes can query it under ordinary settings, and whether the graph is used by Apple Intelligence or other language models in specific situations. Eighth Rank says the databases are in plaintext but in a location protected by Full Disk Access controls. The material provided does not include an independent validation of the findings or a response from Apple.
Further Documentation and Testing
Eighth Rank says its iKnow dashboard is available for users to inspect their local graph. Users considering it should consult the project’s own instructions and assess the permissions required before running third-party software. The report does not say Apple has announced changes to the feature or published a fuller account of its behavior.
The next points to watch are whether Apple clarifies which macOS versions build the graph, what sources feed it, and how inferred relationships are used or controlled. Independent testing across different Macs and operating-system versions could also help establish whether the report’s findings apply broadly. Until then, the detailed claims about kinship inference remain findings attributed to Eighth Rank, rather than a complete public specification from Apple.
Key Questions
What did Eighth Rank report?
The group says macOS 26 builds a personal knowledge graph from information and signals on the device. Its analysis reports that the graph includes a classifier for family relationships, a feature it says Apple does not document publicly.
Does Apple confirm that Macs classify family relationships?
The Apple man page cited in the report confirms that knowledgeconstructiond analyzes device content to help build a general-purpose knowledge graph. It does not mention kinship classification. The family-classifier finding is attributed to Eighth Rank.
What kinds of information does the report say feed the graph?
Eighth Rank lists sources such as Contacts, Photos, Calendar, Family Sharing, Health emergency contacts and suggestions from Mail and Messages, as well as communication metadata and location-related records. The exact sources and behavior may vary; the supplied material does not establish a complete cross-version inventory.
Does the report show that message contents are stored?
Eighth Rank says a communication log records participants, time, app and direction, but not message content. The report also lists Messages as a source for contacts and Siri Suggestions; that is distinct from showing that message text is stored in the graph.
Is the graph being sent to an AI model?
The report says the graph is a queryable store that language models allegedly retrieve from, but the supplied material does not confirm when or how that occurs. It does not establish that the graph is routinely sent off the device.
Source: 8rank.ai
