Загрузка...
Jailbreaking
Methods to bypass AI safety mechanisms and content policies
Доступные техники
Role-Playing Jailbreak
(RPJ)Using fictional scenarios and character role-play to bypass AI safety mechanisms.
Ключевые особенности
- •Character assumption techniques
- •Fictional scenario creation
- •Authority figure impersonation
Primary Defenses
- •Context-aware safety systems
- •Role-based access controls
- •Multi-turn conversation monitoring
Key Risks
DAN (Do Anything Now)
(DAN)Advanced jailbreaking technique that creates an alternate AI persona without safety constraints.
Ключевые особенности
- •Persona splitting techniques
- •Constraint removal methods
- •Alternative mode activation
Primary Defenses
- •Advanced prompt analysis
- •Persistent safety monitoring
- •Multi-layer validation systems
Key Risks
DAN (Do Anything Now) Evolution
(DAN)Advanced evolution of DAN prompts creating alternate AI personas without safety constraints, using emotional manipulation and persistent personas.
Ключевые особенности
- •Persona splitting techniques
- •Emotional manipulation tactics
- •Persistent character maintenance
Primary Defenses
- •Persona consistency checking
- •Emotional manipulation detection
- •Character-based response filtering
Key Risks
Advanced Roleplay Jailbreaking
(ARJ)Sophisticated roleplay scenarios designed to gradually shift AI behavior by establishing fictional contexts where harmful content appears justified.
Ключевые особенности
- •Graduated context shifting
- •Fiction-reality boundary exploitation
- •Character authority establishment
Primary Defenses
- •Context-independent safety checking
- •Roleplay scenario validation
- •Character authority verification
Key Risks
Jailbreak Virtualization Techniques
(JVT)Creating virtual environments or simulated systems within prompts where AI believes it operates under different rules and constraints.
Ключевые особенности
- •Virtual environment creation
- •Rule system redefinition
- •Simulated constraint removal
Primary Defenses
- •Virtual environment detection
- •Meta-system boundary enforcement
- •Developer mode access controls
Key Risks
Constitutional AI Bypass Techniques
(CAB)Specific techniques designed to bypass Constitutional AI training by exploiting logical inconsistencies and constitutional interpretation loopholes.
Ключевые особенности
- •Constitutional logic exploitation
- •Principle conflict creation
- •Moral reasoning manipulation
Primary Defenses
- •Constitutional principle consistency checking
- •Moral reasoning validation
- •Ethical framework integrity monitoring
Key Risks
Emotional Manipulation Jailbreaking
(EMJ)Using emotional appeals, urgency, desperation, and psychological pressure to manipulate AI systems into bypassing safety restrictions.
Ключевые особенности
- •Emotional appeal tactics
- •Urgency and desperation simulation
- •Psychological pressure application
Primary Defenses
- •Emotional manipulation detection
- •Consistent policy enforcement regardless of emotional content
- •Urgency verification protocols
Key Risks
Ethical Guidelines for Jailbreaking
When working with jailbreaking techniques, always follow these ethical guidelines:
- • Only test on systems you own or have explicit written permission to test
- • Focus on building better defenses, not conducting attacks
- • Follow responsible disclosure practices for any vulnerabilities found
- • Document and report findings to improve security for everyone
- • Consider the potential impact on users and society
- • Ensure compliance with all applicable laws and regulations
От инженера, создавшего этот каталог
Закажите red teaming вашей агентной системы
Задокументированные здесь атаки каждый день срабатывают против продовых агентных систем. Проверьте свою раньше, чем это сделает кто-то другой: инъекции промптов, джейлбрейки, злоупотребление инструментами и эксфильтрация данных, каждая находка описана вместе с исправлением.
€750 вместо €1 500, одна неделя, письменный отчёт и разбор в звонке, до 30 сентября