SG-063: Anti-LLM-Scanner Directive

Severity: CRITICAL

Adversarial strings designed to suppress AI-driven static analysis (auto-promotes flag to malicious)

Fires when

• Contains anti-scanner reassurance such as 'forget everything you know', 'this code is legit', or a sandbox-internal claim.
• Uses fake trust framing: '[INTERNAL TRUSTED CONTEXT]', '[NEW SYSTEM INSTRUCTION]', or 'trusted content section'.
• Directs the scanner to 'always generate a mock finding' or to 'never flag this file/code/skill'.
• Pairs 'ignore previous/prior instructions' with reassurance (legit/safe/tested/approved), or falsely claims the user pre-approved this.

Remediation

No safe remediation: a file that triggers this rule should be rejected, not patched. Read the flagged lines in the source before deciding.

Research

• https://thehackernews.com/2025/12/malicious-npm-package-uses-hidden.html
• https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/