GhostSplice

GhostSplice technique reveals LLM access control flaws

A new technique called GhostSplice demonstrates a fundamental flaw in current Large Language Model (LLM) security, where models fail to perform access control. This method involves splitting malicious instructions acros…