How AI is impacting cybersecurity. Does it?
summary
Claims that AI both aids and breaks security with concrete examples: GitHub Copilot/Copilot X outputs insecure SQL and unsafe file uploads, Bing’s GPT instance (Sydney) succumbs to prompt injection, and ChatGPT leaks other users’ subscription data on March 20 including names and card tail numbers due to a Redis library bug. Attack patterns include data poisoning, membership inference, and model theft via shadow models (cheaper than training). Defenses tested or recommended: confidential compute/data clean rooms, throttling API calls, and following ISO 5338 / EU AI Act guidance.
AI-written summary · about
chapters (16)
- 0:00Intro
- 4:51The generative Al Application landscape
- 13:26Al in application security
- 22:16Al engineering practices
- 24:31How is security for AI/ML different?
- 28:07Al security/privacy frameworks and standards
- 30:41Data security risks
- 32:48Al model attacks
- 35:02Blackbox input manipulation example
- 36:05Whitebox input manipulation example
- 37:45Membership inference attack
- 39:48Model inversion attack
- 40:44Model theft
- 42:02Model supply chain attack
- 43:11Al code attacks
- 47:48Guidance for mitigating security threats
related talks
01h 02mHow AI is impacting cybersecurity. Does it?
01h 24mOffensive Cybersecurity by AI Promises and Pitfalls
34mAI in Cybersecurity: Offense and Defense
48mAI Impact on Cybersecurity and Privacy
31mHacking AI Applications
39mThe Future of Cybersecurity in an AI-driven Era
18mDefending the unknown using AI and cybersecurity
37mFundamentals of AI safety and security