Words as weapons: The dark arts of Prompt Engineering
summary
Explains how ChatGPT, custom GPTs, Bing and Co‑pilot can be prompt‑injected to bypass filters, shows retrieving a system prompt and leaking it, and demonstrates hidden prompts in images and white text that make models output “you got hacked.” Shows a CV trick that forces top scores and an invoice attack where a $6,000 bill is made to pay $1,000,000 by a hidden prompt; some jailbreaks and website/robots.txt retrievals work in practice so defenses must change.
AI-written summary · about
chapters (11)
- 0:00Introduction and context
- 1:44Understanding social engineering
- 5:53Guardrails and security
- 9:34System prompts explained
- 12:43Human curiosity and AI
- 17:25Adversarial prompting tactics
- 20:27Real-world exploitation
- 25:55Advanced injection techniques
- 29:49Business automation risks
- 33:44Developer warnings
- 37:16Conclusion and takeaways
related talks
51mPrompt Injection: When Hackers Befriend Your AI
30mBattling the Machine: Cutting-Edge Attacks and Defenses in Generative AI
02h 01mPrompt Engineering and AI Red Teaming
49mUnderstanding Prompt Injection-Techniques,Challenges and Advanced Escalation
22mAI, Make It So - A Dive into the Universe of Prompt Engineering
25mWhen words become attacks: prompt injection through a technical writing lens
45mPrompt-Jacking: The Rise of a New Supply Chain Risk
13mOf course Dave, I'd love to do that - AI Prompt Injection attacks