Keynote - Hacking AI How to Survive the AI Uprising
summary
Walks through concrete attacks and defenses: jailbreaks like copyable DAN prompts and a greedy coordinate gradient exploit that is reported 100% effective against a local model, and a universal black box jailbreaking method with ~85–97% success without model access. Demonstrations include Trojan resume tricks, image perturbation that flips classifier labels, Glaze/Nightshade for poisoning training images, model inversion and Fox face morphing, a three‑sticker Tesla lane hijack, dolphin (ultrasonic) voice injections, and the need for MCP audits and rollback wrappers.
AI-written summary · about
chapters (8)
- 0:00Introduction to AI hacking
- 3:12The history of naive tech
- 6:21Hacking AI text
- 12:47Poisoning AI data
- 14:54Audio and vision vulnerabilities
- 20:36Advanced AI vision attacks
- 29:00Security and future mindset
- 34:36The human advantage
related talks
44mHacking AI - How to Survive the AI Uprising
37mHow to Break AI Systems (Before Someone Else Does)
35mHacking AI Systems: How to (Still) Trick Artificial Intelligence
01h 01mHow to Break AI Systems (Before Someone Else Does)
52mSleeping With One AI Open: An Introduction to Attacks Against...
53mHow to Break AI Systems (Before Someone Else Does)
46mA Short Introduction to Hacking Artificial Intelligence
42mHow to attack AI systems (and how to defend them) !!!!