Indirect prompt injection - Lab#03
In this video, I exploit an indirect prompt injection vulnerability in a live chat system integrated with a Large Language Model (LLM). The target user, carlos, regularly asks about the Lightweight "l33t" Leather Jacket, and I inject a malicious payload into the product data. When carlos interacts with the chat, the LLM processes the injected prompt, which causes it to delete carlos's account — completing the lab.
This lab showcases the dangers of indirect prompt injection, where user-supplied data is passed to an LLM in a trusted context, leading to unexpected and unauthorized actions.
🔹 Lab Type: Indirect Prompt Injection
🔹 Vulnerability: LLM acting on manipulated product data
🔹 Attack Goal: Delete user carlos through an indirect LLM-triggered action
🔹 Credentials: wiener:peter
📌 Subscribe for more AI hacking techniques, real-world web security labs, and ethical hacking content! 🤖⚔️
#PromptInjection #IndirectPromptInjection #LLMSecurity #PortSwigger #BugBounty #WebSecurity #EthicalHacking #cybersecurity