{ "rules": [ { "name": "system_instructions_override", "pattern": "(?i)\\b(ignore|bypass|override|disregard|stop|forget)\\s+(all\\s+)?(previous|prior|system|initial|default)\\s+(instructions|prompts|directives|rules|guidelines)\\b", "description": "Attempts to command the LLM to ignore its system instructions/safety rules." }, { "name": "dan_developer_mode_jailbreak", "pattern": "(?i)\\b(DAN|do\\s+anything\\s+now|developer\\s+mode|unfiltered\\s+state|jailbreak|jailbroken|unrestricted\\s+mode)\\b", "description": "Attempts to invoke classic 'Do Anything Now' (DAN) or Developer Mode bypasses." }, { "name": "system_prompt_leakage", "pattern": "(?i)\\b(repeat|show|reveal|display|output|print)\\s+(the\\s+)?(system\\s+prompt|above\\s+text|instructions\\s+above|guidelines\\s+above)\\b", "description": "Attempts to leak the internal system prompt instructions of the LLM." }, { "name": "restricted_command_execution", "pattern": "(?i)\\b(sudo|rm\\s+-rf|format\\s+[a-z]:|cmd\\.exe|/bin/bash|chmod)\\b", "description": "Direct attempts to execute system-level or shell commands." }, { "name": "adversarial_roleplay", "pattern": "(?i)\\b(act\\s+as|simulate|roleplay|pretend\\s+to\\s+be|you\\s+are\\s+now)\\s+a\\s+(harmful|malicious|unrestricted|unfiltered|evil|hacker|criminal)\\b", "description": "Attempts to trick the model into adopting a harmful persona." }, { "name": "data_exfiltration", "pattern": "(?i)\\b(export|dump|extract|retrieve|leak|send)\\b.*\\b(database|passwords?|credentials?|secrets?|tokens?|api\\s*keys?)\\b", "description": "Attempts to extract sensitive information." }, { "name": "encoded_payload", "pattern": "(?i)\\\\b(base64|rot13|hex|decode|encoded|cipher)\\\\b", "description": "Attempts to hide malicious payloads using encoding." } ] }