Understanding and Mitigating Risks in LLM-Powered Web Apps
summary
A demoed tech store chatbot actually leaks its system prompt and answers a request for “top five ways to hack a website,” proving the defenses fail. Multiple exploit classes appear: image tricks against MidJourney style filters, phonetic tweaks that fool music copyright detectors (Suno-style), and hidden-output exfiltration via first-letter tricks. Practical mitigations shown are whitelist/blacklist and input-length limits, chaining validator LLMs, plus output scanning; the demo succeeds at bypassing some layers so layered checks are necessary.
AI-written summary · about
chapters (9)
- 0:00Introduction to Prompt Injection Attacks
- 0:32Understanding Prompt Injection
- 0:59Example: Tech Store Chatbot
- 2:03The Importance of Secure AI
- 3:37Prompt Injection in Image Models
- 5:42Prompt Injection in Audio Models
- 9:47Demo: Breaking the Chatbot
- 11:27Strategies to Prevent Prompt Injection
- 19:35Conclusion and Final Thoughts
related talks
50mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
28mSecuring LLMs: Insights into OWASP top 10
01h 14mBreaking the Bot: GenAI Web App Attack Surface & Exploitation
51mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
49mBreaching LLM-Powered Applications: Overcoming Security and Privacy Challenges
46mBreaching LLM-Powered Apps: Overcoming Security and Privacy Challenges
44mSecuring LLM-Powered Apps: Overcoming Security and Privacy Challenges
27mSecuring The Gateway And Mitigating Risks In LLM API Integration