Every Mac comes with an (undocumented) family member classifier

Recent research claims that modern versions of macOS run undocumented background services that build a detailed “personal knowledge graph” of users and their contacts: who they are, how they communicate, where they go, and inferred family or social relationships, drawn from email, messages, photos, and other apps. Commenters are alarmed that this appears to happen even with Siri and Apple Intelligence disabled, is not clearly documented, cannot easily be turned off, and may raise GDPR and broader privacy concerns despite being stored locally. Some see potential utility for on-device assistants, but many argue that such profiling should be opt‑in, transparent, and either more strongly protected or avoided altogether, with a number of users reconsidering macOS in favor of Linux or other alternatives.

Scope and Behavior of the Mac “Knowledge Graph”

  • Macs since 2022 run daemons that build a personal knowledge graph under `/Library/IntelligencePlatform`.
  • Data sources include contacts, messages, calls, calendars, photos (faces and co‑occurrence), app usage, and even data synced from iPhones and other Macs via iCloud.
  • It’s stored in plaintext SQLite databases but gated by macOS’s Full Disk Access (TCC). Apple Intelligence, Siri, Photos, Messages, etc. can query it.
  • Some users see rich, cross‑device graphs (including work Slack, Signal contacts, managed profiles); others see almost nothing or hilariously wrong inferences.

Privacy, Consent, and Legal Questions

  • Many find the feature “horrifying,” especially because:
    • It runs even with Siri/Apple Intelligence disabled.
    • There is no obvious UI, documentation, or opt‑out.
  • Several argue it likely violates GDPR and specific German law (TDDDG §25) due to repurposing personal data without explicit consent.
  • Others counter that data is processed locally on the user’s machine, so Apple may not be a data controller in the GDPR sense; debate remains unresolved and “unclear.”

Security and Access Concerns

  • Any app granted Full Disk Access (including terminals, debuggers, or tools) can read and exfiltrate the graph.
  • Extra risk if people view it via a browser with malicious extensions; some say this should be documented explicitly.
  • Worries extend to law enforcement, hostile governments, employers, and litigants using the graph as a high‑value dossier.
  • Side thread: claims that macOS ignores user DNS settings pre‑boot to contact Apple security domains, with discussion of Pi‑hole, DoH/DoT, and router‑level defenses.

Disabling or Mitigating

  • Consensus: there is no supported way to turn it off.
  • Suggestions include:
    • Killing/permission‑stripping the daemons (likely undone by updates/SIP).
    • Locking ~/Library/IntelligencePlatform with immutable flags and root ownership.
    • Avoiding granting Full Disk Access to untrusted apps.
    • More radically, switching to Linux/BSD or using Macs only offline/air‑gapped.

Usefulness vs. Creepiness

  • Proponents: such a graph is needed for natural commands like “text James” or “call my boss” and is preferable to cloud profiling if it truly stays local.
  • Critics: concentration of detailed relationship data is inherently dangerous; benefits don’t justify silent collection without consent or controls.

Reaction to the iKnow Tool and AI‑Written Prose

  • The inspection tool is appreciated for surfacing the graph, but many dislike that its interface text and much of the article are LLM‑generated.
  • Common complaints: verbose, unclear, “AI slop” style that undermines trust in otherwise strong technical findings.