2 matches found
AgentWatcher
AgentWatcher AgentWatcher è una difesa basata sul rilevamento contro l'iniezione indiretta di prompt negli agenti LLM. Innanzitutto esegue l'attribuzione causale del contesto sul contesto non attendibile per individuare i contesti più influenti, quindi applica un LLM monitor che classifica tali...
5.4AI score
SaveExploits0References5
AgentWatcher: A Rule-Based Prompt Injection Monitor
Large language models LLMs and their applications, such as agents, are highly vulnerable to prompt injection attacks. State-of-the-art prompt injection detection methods have the following limitations: 1 their effectiveness degrades significantly as context length increases, and 2 they lack...
5.9AI score
SaveExploits0
20