LLM Security 101 - An Introduction to AI Red Teaming
summary
Live demo shows an automated red team tool extracting a hidden password from a defended model, using Pirate (Python Risk Identification Tool) to iterate prompts and score leaks. A concrete exploit uses Google AI Studio image-markdown exfiltration where a malicious text file causes confidential data to be fetched and leaked. Claims: automation rapidly amplifies social‑engineering strategies, jailbreaks and indirect prompt injections work in practice, and simple mitigations like rate limits, read‑only access and LLM watchdogs materially reduce risk.
AI-written summary · about
chapters (8)
- 0:00LLM security basics
- 3:19Security concepts and threats
- 7:58Prompt injection attacks
- 11:54Indirect prompt injection
- 16:11Red teaming framework
- 18:15Demo: Pirate tool
- 25:01Next steps in LLM security
- 29:14Q&A
related talks
01h 01mIntroduction to AI Security
26mWhen AI Gets Manipulated: Security Risks in LLM Applications
33mSecuring Agentic AI With PyTorch: Threat Modeling & LLM Red Teaming in Practice
30mNavigating the New AI Security Landscape
57mAI Applications Security Puzzle
20mAI Red Teaming and Automation: Exploring Societal Risks in...
43mSecurity of GenAI & LLMs: A New Era of Ethical Hacking
01h 24mOffensive Cybersecurity by AI Promises and Pitfalls