RemoveMacAI CLI Tool Disables Apple Intelligence and Deletes On‑Device AI Models on macOS 27

RemoveMacAI CLI Tool Disables Apple Intelligence and Deletes On‑Device AI Models on macOS 27

What Happened – An open‑source command‑line utility, RemoveMacAI, was released on GitHub. When executed on Apple‑silicon Macs running macOS 27, it installs a configuration profile that disables Apple Intelligence features and deletes roughly 12 GB of locally stored AI models. The tool also blocks future model downloads while keeping System Integrity Protection enabled.

Why It Matters for Trust & Control Assurance

• Demonstrates the need for documented configuration‑change controls and evidence that AI model usage aligns with corporate policy.
• Highlights a gap where on‑device AI assets can be removed without centralized oversight, complicating audit trails for AI governance.
• Provides a concrete example of a control‑mapping scenario: mapping “AI model lifecycle management” to a single VCF control objective satisfies multiple framework requirements (e.g., NIST AI RMF, ISO 42001).

Who Is Affected – Enterprises and individual users that have enabled Apple Intelligence on macOS 27, especially regulated sectors that rely on on‑device AI for data‑sensitive workflows (e.g., finance, healthcare, legal).

Recommended Actions

• Inventory all macOS endpoints for enabled Apple Intelligence features and record model storage locations.
• Align your AI model‑lifecycle policy with a VCF control objective (e.g., “Manage and monitor AI model acquisition, deployment, and retirement”).
• Capture configuration‑profile changes as immutable evidence in your continuous monitoring platform.
• Test the revert command in a controlled environment to verify that model re‑download restores expected functionality without violating policy.

Source: Help Net Security

Technical Notes

• The tool runs a single curl or Homebrew command, prompts for profile approval, and writes Apple restriction keys to a local port, causing model fetch failures.
• No network traffic is generated by the tool itself; it is MIT‑licensed and auditable.
• A dry‑run mode shows intended changes; a flag allows selective feature retention.
• System Integrity Protection remains enabled; no /System files are altered.

Source: same as above