What Are the Security Risks Associated with AI-Generated Code? 🤔
summary
Live demo builds a small app with GitHub Copilot then finds and exploits real issues in the generated code, showing concrete failures: prompt injection that leaks secrets (Gandalf challenge and a Chevrolet bot case), insecure output handling, denial of service from missing max iterations in LangChain, and supply‑chain leaks when internal code is pasted to public AI. The session names exact tools, reproduces attacks, and gives practical mitigations — input validation, output sanitization, least privilege, human oversight and monitoring — so viewers know what to test and fix.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:30Securing AI
- 2:14Three Aspects of AI
- 3:46OASP Top 10 for LTM
- 5:07Prompt Injection
- 8:56Gendal
- 10:45Excessive Agency
- 16:06Training Poisoning
- 20:03Overreliance
- 22:40Is AI Reliable
- 24:43The Problem
- 26:42The Code
related talks
46mBreaking AI: Live coding and hacking applications with Generative AI
58mSecuring Code in the Age of AI
46mBreaking AI: Live coding and hacking applicattions with Generative AI
28mAI-Generated Code: Unmasking the Security Pitfalls
31mHacking AI Applications
41mCode Security Reinvented Navigating the era of AI
26mSmarter Code, Sneakier Risks: Supply Chain Security in the Age of AI
44mBreaking AI: Live Coding and Hacking Applications with Generative AI