Using GenAI on your code, what could possibly go wrong?
recording made private on YouTubehttps://youtu.be/krDJlrw5mM0
summary
Academic benchmarks find roughly 40% of generated snippets contain vulnerabilities (2021–2023 studies) and code velocity rises as developers accept LLM output. Demos use Semantic Kernel + Azure OpenAI where an available file-access plugin lets the model read /etc/hosts — a concrete semantic kernel demo exfiltrated /etc/hosts. Other realistic risks: prompt‑injection and jailbreaks, plugins/agents chaining actions, data‑poisoning where altering ~1% of training data can backdoor behavior, and gaps in fixing large refactors like crypto replacements.
AI-written summary · about
related talks
55mUsing GenAI on and inside your code, what could possibly go wrong?
52mUsing GenAI on and inside your code, what could possibly go wrong?
31mHacking AI Applications
30mWhat Are the Security Risks Associated with AI-Generated Code? 🤔
12mTrusting AI with Code: How Secure is AI-written code?
46mBreaking AI: Live coding and hacking applications with Generative AI
01h 24mOffensive Cybersecurity by AI Promises and Pitfalls
43mSecurity of GenAI & LLMs: A New Era of Ethical Hacking