Initial Access, within the context of LLM security, represents the first stage of an attack where an adversary gains control or influence over an LLM-based system, primarily through prompt injection. It is the foundational step in a multi-stage 'promptware' kill chain, analogous to traditional malware campaigns.
Initial Access in LLM security refers to the first step an attacker takes to gain control over an AI system, typically by using clever text prompts. This initial breach is crucial because it sets the stage for more serious attacks, much like how hackers first get into a computer network.
prompt injection, adversarial prompting, input manipulation
Was this definition helpful?