Attack, Defense & Danger in the Age of AI
Talkat Cloud Native + Kubernetes AI Day NA 2024 · on Tuesday, 2024-11-12· at 09:45· Salt Palace | Level 1 | Grand Ballroom A·
summary
Live demo runs a Kubernetes workload called Whopper GPT that scales a tic‑tac‑toe deployment to 50 and uses a discovered system prompt token to control replicas, illustrating a working prompt injection exploit that exfiltrates credentials. Shows automated typo-squatting (pip package name traps like hugging face‑CLI/math plot lib) and simple AI-driven obfuscation and minification to hide backdoors. Claims widespread exploit generation by low‑skill actors, explains why signature detection fails, and recommends layered LLM filters, proxies, OpenSSF scorecards, deobfuscation, and human review.
AI-written summary · about
chapters (9)
- 0:00AI risks and opportunities
- 2:27Demo: Whopper GPT arsenal
- 4:11Name squatting and supply chain risks
- 8:27AI exploit generation
- 10:13Obfuscation attacks with AI
- 13:07Prompt injection in AI
- 18:28Defensive strategies
- 21:58Secure prompts and controls
- 23:46Testing defenses with Whopper GPT
related talks
01h 24mOffensive Cybersecurity by AI Promises and Pitfalls
53mHow to Break AI Systems (Before Someone Else Does)
30mBattling the Machine: Cutting-Edge Attacks and Defenses in Generative AI
36mAgentic AI Under Attack: Live Demos of Exploits Through Autonomy and Trust
25mPreventing an OWASP Top 10 in the world of AI
31mHacking AI Applications
37mHow to Break AI Systems (Before Someone Else Does)
46mBreaking AI: Live coding and hacking applications with Generative AI