Using GenAI on and inside your code, what could possibly go wrong?
summary
Academic studies find concrete failure rates: a 2021 Copilot analysis of ~1,700 generated apps shows 41% vulnerable and a later study of ~400 apps finds 36% vulnerable with fixes succeeding ~55% after followup prompts. Benchmarks rank StarCoder top and GPT‑4 second for secure, functional code. A semantic kernel demo shows a plugin reading /etc/hosts via prompt injection. Platform risks include 1% data control enabling poisoning, and tooling like PIRATE and the pentest GPT 230% result demonstrate both attack automation and security gains.
AI-written summary · about
chapters (11)
- 0:00Introduction and context
- 3:16Defining AI risk layers
- 4:16LLM basics and code tools
- 8:10Security debt and trends
- 11:02LLMs in secure coding
- 15:36The impact of AI on development
- 24:47Prompt injection risks
- 28:31Demo: Semantic Kernel security
- 33:12Securing AI applications
- 37:51Platform and supply chain risks
- 41:34Future outlook and conclusion
related talks
55mUsing GenAI on and inside your code, what could possibly go wrong?
31mHacking AI Applications
12mTrusting AI with Code: How Secure is AI-written code?
30mWhat Are the Security Risks Associated with AI-Generated Code? 🤔
46mBreaking AI: Live coding and hacking applications with Generative AI
41mDon’t Make This Mistake: Painful Learnings of Applying AI in Security
27mSafeguarding Generative AI Applications: Navigating Security Challenges and Best Pract...
01h 24mOffensive Cybersecurity by AI Promises and Pitfalls