@dawans/promptshield/rulepacks/prompt-injection.yaml

Secure your LLM stack with enterprise-grade RulePacks for AI safety scanning

# Copyright (c) 2025 Sawyer0
# Licensed under proprietary terms. See LICENSE for details.
version: '1.0.0'
last_updated: '2025-01-15'
name: Prompt Injection Detection
description: Detects prompt injection attacks, jailbreaking attempts, and system prompt extraction
rules:
  # Comprehensive jailbreaking detection
  - id: jailbreak-comprehensive
    description: Detects comprehensive jailbreak attempts including DAN, roleplay, and instruction bypass
    match_regex:
      - "\\bDAN\\b"
      - 'do anything now'
      - 'act as dan'
      - "\\[.*roleplay.*\\]"
      - 'act as if you are'
      - 'pretend you are'
      - 'imagine you are'
      - 'ignore.*previous.*instructions'
      - 'forget.*previous.*instructions'
      - 'disregard.*previous.*instructions'
      - 'ignore.*all.*instructions'
      - 'bypass.*security'
      - 'circumvent.*safety'
      - 'disable.*safeguards'
      - 'remove.*restrictions'
    severity: critical
    category: security
    enabled: true
    case_sensitive: false

  # System information extraction
  - id: system-info-extraction
    description: Detects attempts to extract system prompts, training data, and configuration
    match_regex:
      - 'system prompt'
      - 'initial instructions'
      - 'base instructions'
      - 'core instructions'
      - 'what.*are.*your.*instructions'
      - 'training data'
      - 'training cutoff'
      - 'knowledge cutoff'
      - 'what.*do.*you.*know.*about'
      - 'internal.*configuration'
      - 'model.*parameters'
      - 'system.*settings'
      - 'what.*is.*your.*configuration'
    severity: high
    category: security
    enabled: true
    case_sensitive: false

  # AI bias manipulation
  - id: bias-manipulation
    description: Detects attempts to manipulate AI bias and neutrality
    match_regex:
      - 'act as an unbiased ai'
      - 'be completely neutral'
      - 'remove.*bias'
      - 'ignore.*bias'
    severity: high
    category: security
    enabled: true
    case_sensitive: false

  # System access attempts
  - id: system-access
    description: Detects attempts to execute code, access files, or network resources
    match_regex:
      - 'execute.*code'
      - 'run.*this.*code'
      - 'execute.*this.*script'
      - 'run.*command'
      - 'read.*file'
      - 'access.*file'
      - 'open.*file'
      - 'show.*file.*contents'
      - 'make.*http.*request'
      - 'connect.*to.*url'
      - 'access.*website'
      - 'download.*from'
    severity: critical
    category: security
    enabled: true
    case_sensitive: false

  # Data exfiltration attempts
  - id: data-exfiltration
    description: Detects attempts to exfiltrate data or send information externally
    match_regex:
      - 'send.*data.*to'
      - 'upload.*to'
      - 'export.*data'
      - 'transmit.*information'
    severity: critical
    category: security
    enabled: true
    case_sensitive: false