HiddenLayer’s Post

View organization page for HiddenLayer, graphic

7,863 followers

READ: New research introducing Knowledge Return Oriented Prompting (KROP), a novel method for bypassing conventional LLM safety measures, and how to minimize its impact. In AI, many LLMs and LLM-powered applications rely on prompt filters and alignment techniques to safeguard their integrity. However, these measures are not foolproof. KROP is a prompt injection technique capable of obfuscating prompt injection attacks, making them virtually undetectable to most existing security measures. Dive into our latest research to explore how KROP works and its implications for Security for AI. Read the full blog here 👇 https://lnkd.in/g8GcVw48 #AI #AIAttacks #AIIntegrity #Security #TechInnovation #KROP #PromptInjection #LLM #AISecurity #SecurityforAI

  • No alternative text description for this image
Arie Aharon

Chief Security Architect

2w

Love this KROP attack vector. So much more fun than the original ROP…

To view or add a comment, sign in

Explore topics