Application-Aware Layer-7 Security Framework for Cloud APIs Using...
summary
A framework uses Envoy sidecars plus WebAssembly plugins to call LLMs for per-endpoint LLM intent tagging and extract intent parameters, then evaluates policies in Open Policy Agent. Intent-tagging accuracy ranges 90–99% (user-registration ~99%, “none” ~85%). LLM calls spike latency (requests rise from ~400 ms to ~3 s) but endpoint-level caching amortizes cost. Demo enforces a cart-quantity limit of 10 across different apps (Shopiser, Juice Shop) and correctly denies requests with quantity 11, showing the policy enforcement pipeline works end-to-end.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 2:32Cloud API security challenges
- 7:35API usage security risks
- 9:01Solution technologies
- 11:14AI for request intents
- 15:12Intent tag relevance across apps
- 16:59Intent parameters extraction
- 19:51Intent classification prompts
- 23:24LLM parameter extraction
- 24:09OPA and WebAssembly policies
- 25:37Intent tagging accuracy
- 25:59Policy enforcement demo
- 29:40Research insights and limits
- 31:09Latency and caching issues
- 33:46Q&A
related talks
36mDynamic and Context-Aware Security Policies for the Cloud Using...
38mPhishing LLMs: Reeling in the Machine
30mSecuring LLMs in Kubernetes: Best Practices
27mSecuring The Gateway And Mitigating Risks In LLM API Integration
47mA Step-by-Step Guide to Securing Large Language Models (LLMs)
21mAI-Powered O11y with LLMs: Resilience in Cloud PaaS Archs
20mSecurity Threats in Modern LLM Applications
38mBehind the Firewall: Designing Secure LLM Pipelines for Regulated Data