AI Under the Hood: Unmasking Hidden Threats
summary
A meta-analysis of 14 AI threat-modeling frameworks yields a unified library that collapses ~429 listed threats into a deduplicated set, converted into roughly 47 questions for manual threat modeling and grouped into Baseline, Continuous Learning, and User Data interaction tracks. The project is open source on GitHub as guardrail, already cited by OAS Project Top 10 NIST EI RMF playbook and OECD catalogs, and plans to move from questionnaire-driven checks to automated adversarial testing tools.
AI-written summary · about
chapters (12)
- 0:00AI security threats intro
- 1:05Motivation for securing AI
- 3:32Challenges in AI security
- 6:05Offensive vs defensive AI threats
- 9:38Steps to secure AI apps
- 13:30End-to-end AI security
- 15:37Guardrail threat library
- 20:21Guardrail threat modeling
- 22:05Guardrail progress visibility
- 23:08Future adversarial testing
- 24:17Guardrail project demo
- 26:29Q&A
related talks
25mThreat Modeling for AI Apps with Attacks as Code
18mCommit Secure AI: Embedding Security in Code, Pipelines, and Production
15mAI Security Simplified: Turning Complexity into Cohesion
15mCyber Risk Assessment, IT Asset Management and… AI obviously
41mThreat Modelling for ML/AI systems
33mSecuring Agentic AI With PyTorch: Threat Modeling & LLM Red Teaming in Practice
34mInvisible Infiltration of AI Supply Chains: Protective Measures from Adversari...
53mAI-equipped Threat Actors Versus AI-enhanced Cyber Tools: Who Wins?