Explore a 13-minute video presentation examining security vulnerabilities in Large Language Model (LLM) browser agents, focusing on prompt injection risks detailed in WithSecure Labs' research. Learn through a practical attack demonstration showing mailbox information exfiltration, understand the operational mechanics of LLM browser agents using TaxyAI as an example, and discover key attack vectors along with their limitations. Gain valuable insights into defensive strategies, including agency limitation and detection methods, to protect against these emerging security threats. Access a comprehensive mindmap for visual reference and follow along with timestamped segments covering the complete analysis from introduction through defense mechanisms.
Security Risks of LLM Browser Agents - Understanding Prompt Injection Vulnerabilities
Overview
Syllabus
- - Introduction
- - Attack Demo Exfiltrate Information from Mailbox
- - How LLM Browser Agents Work TaxyAI Operational Loop
- - Injection Attack
- - Attack Caveats and Limitations
- - Defence Strategies Limit Agency, Detection
Taught by
Donato Capitella
Related Courses
-
LLM Pentesting: Mastering Security Testing for AI Models
-
Hacking and Securing LLM Applications - Understanding Browser Control Security Risks
-
Building LLM Applications Securely - Understanding Risks and Mitigation Strategies
-
Prompt Injection Attacks and Security in LLM-Powered Agents - ReAct and Langchain
-
Ethical Hacking against and with AI/LLM/ML (Lite Version!)
-
Multi-Chain Prompt Injection and Jailbreaking in LLM Applications - Security Testing and Defense Strategies
