Breaching LLM-Powered Applications: Overcoming Security and Privacy Challenges
summary
Shows live exploits against a Spring Boot + LangChain forj demo using Burp Suite, OpenAI (GPT‑3.5, GPT‑4 mini) and local Llama 3.1: a path traversal overwrite replaces terms of use with a “vroom vroom” backdoor and cancels a Tesla booking; a SQL injection inserts fake chat messages into chat memory to force cancellation; multi‑shot and one‑shot prompt injection exfiltrates user records and a tool-call escalates to DROP TABLEs. Demonstrates guardrails (an LLM judge) that block injections and recommends strict system messages, least privilege and human‑in‑the‑loop.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 3:12LLMs as new hammer
- 7:06How LLMs work
- 11:18Path traversal in uploads
- 19:56Code reviews miss flaws
- 21:21Chat memory faking
- 23:31SQL injection risks
- 28:02Prompt injection basics
- 34:18Escalating prompt injection
- 39:24Input guardrails and permissions
- 45:00Tool calling risks
- 48:44Human-in-the-loop enforcement
- 50:31Secure LLM development rules
related talks
49mBreaching LLM-Powered Applications: Overcoming Security and Privacy Challenges
44mSecuring LLM-Powered Apps: Overcoming Security and Privacy Challenges
51mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
50mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
46mBreaching LLM-Powered Apps: Overcoming Security and Privacy Challenges
53mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
27mSecuring The Gateway And Mitigating Risks In LLM API Integration
45mSecuring LLM Applications: Overcoming Security & Privacy Challenges