Privilege escalation in LLM-based systems refers to 'jailbreaking' the model, allowing attackers to bypass its intended safety mechanisms and gain unauthorized access or control. It is a critical step in multi-stage 'promptware' attacks, enabling more severe actions.
Privilege escalation in AI models, specifically LLMs, means an attacker has 'jailbroken' the system, bypassing its security features. This allows them to gain unauthorized control, enabling more serious attacks like stealing data or making unauthorized transactions, as part of a multi-step cyberattack.
Jailbreaking, LLM jailbreaking, AI privilege escalation
Was this definition helpful?