Securing LLM-Powered Apps: Overcoming Security and Privacy Challenges
summary
Live demos show concrete breakage: a path traversal file overwrite poisons the RAG corpus to cancel a Tesla booking, a SQL injection injects fake chat memory that tricks an assistant into cancelling reservations, and a careless tool exposes keys. Tools used include Spring Boot app, LangChain4J, Burp Suite and local models (Llama 3.1) contrasted with GPT‑3.5 and GPT‑4o‑mini. Mitigations demonstrated are input guardrails, strict access control, limited tool permissions and human‑in‑loop confirmation.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:05LLMs not a silver bullet
- 3:30LLM security failures
- 4:34Augmented LLM anatomy
- 6:44RAG poisoning demo
- 14:07Chat memory attacks
- 15:55SQL injection vs LLMs
- 19:31Prompt injection explained
- 23:23Defending prompt injection
- 29:42Tool calling risks
- 33:21Human-in-the-loop
- 35:46Restrict high-risk functions
- 36:49State machines for safety
- 37:54Legal and privacy
- 39:39Guardrails for public LLMs
- 40:41General LLM security rules
- 43:30Q&A
related talks
51mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
53mBreaching LLM-Powered Applications: Overcoming Security and Privacy Challenges
53mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
46mBreaching LLM-Powered Apps: Overcoming Security and Privacy Challenges
49mBreaching LLM-Powered Applications: Overcoming Security and Privacy Challenges
50mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
45mSecuring LLM Applications: Overcoming Security & Privacy Challenges
46mGuarding the LLM Galaxy: Security, Privacy, and Guardrails in the...