Harnessing LLMs for Next-Level Security Analysis
summary
Claims that LLMs can automate threat modeling end to end by mining source code to build architecture models, extract use cases, and surface threats, backed by interviews with 40+ organizations and demos on an eShop reference app. The prototype uses multi‑pass prompting, diagrams‑as‑code (mermaid), GPT‑style models (DPT/GP4 referenced), and categorization to reduce hallucination, producing generated diagrams in minutes and suggested PRs for fixes; limitations include token limits, inconsistent outputs, and the need for QA.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:27Security by Design challenges
- 7:14Architecture modeling accuracy
- 12:57Diagrams as code with LLMs
- 16:10LLMs for security analysis
- 20:55LLM data extraction
- 25:58LLMs and Labradors traits
- 30:18Threat identification with LLMs
- 37:05Countermeasures with LLMs
- 42:27Security steps to code
- 45:23Future of automated security
- 46:11Q&A
related talks
39mThreat Modeling Using LLMs
50mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
39mLLMs everywhere: The future is now and the past keeps repeating itself
23mSecurity and auditing tools in Large Language Models (LLM)
47mA Step-by-Step Guide to Securing Large Language Models (LLMs)
32mThe Sorcerer’s Apprentice: The craft of building reliable solutions with LLMs- 39mThreat Modeling Using LLMs
44mI Abuse Your LLM Live with 10 Security Vulnerabilities