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
An Eighth Rank report says macOS 26 builds an on-device knowledge graph from information and activity signals across a Mac, with classifications that can represent relationships between people. Apple’s public documentation describes the graph only in broad terms, and the report’s findings have not been independently verified here.
A report by security research group Eighth Rank says macOS 26 builds an on-device knowledge graph that can connect a Mac’s owner with people, places, events and patterns of activity, including kinship or family relationships. The report describes the system’s databases and data sources, but Apple’s public documentation does not spell out the graph’s full contents or explain how relationship inferences are made.
Eighth Rank says two macOS background processes, knowledgeconstructiond and intelligenceplatformd, work together to build the graph. The report examined a Mac running macOS 26.6.2 and found a collection of SQLite databases in the user’s IntelligencePlatform directory. It says the databases include graph.db, a store organized around entities and relationships, with each relationship carrying information such as confidence, source and timestamp.
The analysis describes a graph vocabulary of 181 entity classes and 278 predicates. According to Eighth Rank, the system draws on sources including Contacts, Photos, Calendar, Family Sharing, Health’s Medical ID, Siri Suggestions, and information provided by apps. Separate tables reportedly contain communication metadata, location and movement information, and device signals. The report says its communication log records who communicated, when, through which app and in which direction, but not message content.
Eighth Rank also describes a data stream called ContextualUnderstanding.PhysicalSocial, whose record format includes people’s names, phone numbers or email addresses alongside start and end times. The researchers say that format appears to record who may have been physically near the user. That interpretation is based on their examination of the data structure; the source material does not establish exactly how Apple uses each signal or how accurate any resulting inference is.
A More Personal Device Profile
A graph that links contacts, communications, locations and shared events can make on-device services more context-aware, but it also concentrates sensitive details about a user’s relationships and routines. The concern raised by Eighth Rank is not simply that a Mac stores contacts or calendar entries: it is that the system may combine separate records into inferences about how people are connected.
That matters for privacy and security because a consolidated profile could expose information that is less obvious when each source is viewed separately. Eighth Rank frames the graph as a potential privacy single point of failure if an attacker or other unauthorized party gained access to it. The report presents that as a risk scenario, not evidence that the graph has been accessed or misused.
The report says the databases are stored in plaintext within a directory protected by Full Disk Access controls. That description should not be read as meaning the files are freely accessible to any app: the source says access is restricted by macOS privacy protections. The practical protections, threat scenarios and behavior under different settings are not fully established by the material provided.
Popular right now
Breville Bambino Espresso Machine, Brushed Stainless Steel
Breville BES870XL Barista Express Espresso Machine
Bose QuietComfort Wireless Noise-Cancelling Headphones, Cypress Green
As an affiliate, we earn on qualifying purchases.
What Apple Documents Publicly
Apple’s man page for knowledgeconstructiond describes it as a daemon that analyzes on-device content to build a general-purpose knowledge graph representing the device and its user. It says the process works with intelligenceplatformd, which handles some heavier processing. That description confirms the broad purpose of the service, but it does not list all data sources or document the family-relationship classifications discussed by Eighth Rank.
The report notes that Apple researchers published work on graph-construction techniques in 2023, while arguing that the integration into macOS remains sparsely documented. It also points to earlier independent examinations: Nic Lydon wrote about finding a personal knowledge graph on a Mac, and forensic researcher Alexis Brignoni described examining a related store on iOS. Eighth Rank’s detailed analysis is specifically based on macOS 26.6.2; findings from one operating-system version do not by themselves establish identical behavior across every Mac or release.
“A daemon that analyzes content on the device to build a general purpose knowledge graph.”
— Apple’s macOS man page for knowledgeconstructiond
How Accurate Are Its Inferences?
The report’s findings are an analysis by Eighth Rank, not a full technical description published by Apple. The material does not establish which macOS versions contain each component, whether every listed data source is used on every device, or how users’ settings affect collection and processing. It also does not explain the criteria or accuracy of the graph’s kinship classifications.
The report’s description of the physical-social stream is an interpretation of its record format; the material does not confirm that it is a definitive log of everyone physically present. Nor does the source establish that the graph is queried by large language models, despite raising that possibility in its introduction. The report’s stated security thought experiment is not evidence of a breach, and no misuse is identified in the supplied material.
More Documentation and Testing Needed
Eighth Rank says it released iKnow, a local dashboard intended to let Mac users inspect their personal knowledge graph. The source material does not provide details about the tool’s compatibility, independent review or the safeguards users should apply when examining sensitive records.
Further testing across macOS releases and configurations would help establish which records are created, how long they remain, what controls affect them and how relationship classifications are generated. Apple has not provided a response in the source material. Until more information is available, the report supports a narrower conclusion: macOS has a documented on-device knowledge-graph service, while the reported breadth of its data and its family inferences remain incompletely explained in public documentation.
Key Questions
What did Eighth Rank find on Macs?
Eighth Rank says macOS 26.6.2 stores a personal knowledge graph in a set of SQLite databases and that the graph can represent people and relationships. The findings come from the group’s examination of that operating-system version.
Does the graph read message content?
The report says its communication-log table contains metadata such as who communicated, when, through which app and in which direction, but not message content. It separately says the system can draw on information such as contacts and suggestions from Mail and Messages; the full processing details are not clear from the source material.
Has Apple confirmed it classifies family members?
Apple’s cited man page confirms that a daemon builds a general-purpose knowledge graph, but it does not describe kinship inference. The family-relationship finding is reported by Eighth Rank and is not confirmed in the supplied material by Apple.
Is there evidence the databases were accessed by an attacker?
No. Eighth Rank discusses a possible security risk if a consolidated personal graph were accessed by an unauthorized party, but the source material does not report a breach or misuse.
Can users inspect their graph?
Eighth Rank says it released a local dashboard called iKnow for inspecting the graph on a Mac. The supplied material does not detail its compatibility or provide an independent assessment of the tool.
Source: 8rank.ai
