Inside AI Security with
summary
Overview of concrete AI security work: classifies attacks as backdoors and data poisoning (research shows 1% poisoning can flip classifiers), demoed a fingerprint backdoor that makes an esoteric phrase have ~100% probability versus 10^-8 baseline, and describes model theft, inferential extraction, and adversarial misclassification. Describes defenses: Prompt Shields/Prompt Shield, Azure Content Safety, Defender integrations, hidden-layer model scanning, and red‑team tooling (PIRATE, Crescendo automation, Master Key findings) plus practical plugin and data‑classification precautions.
AI-written summary · about
chapters (9)
- 0:00AI security fundamentals
- 3:28Responsible AI governance
- 7:00AI model platform risks
- 15:55Securing AI apps and data
- 23:03Prompt injection attacks
- 30:54Jailbreaking AI systems
- 39:27Content filtering and monitoring
- 42:43Red teaming AI security
- 45:57Key takeaways
related talks
18mCommit Secure AI: Embedding Security in Code, Pipelines, and Production
46mA Short Introduction to Hacking Artificial Intelligence
39mThe Future of Cybersecurity in an AI-driven Era
45mSecuring AI: There Is No Try, Only Do!
49mThe Intersection of AI and Cybersecurity
37mKeynote - Hacking AI How to Survive the AI Uprising
35mGenAI Application Security: Not Just Prompt Injection
45mHardening AI/ML Systems - The Next Frontier of Cybersecurity