Building AI Security In: MLSecOps in Practice
summary
Overview of where attackers actually target ML pipelines: a demonstrated MLflow vulnerability takeover lets adversaries alter models and siphon value, and Synopsys research shows 80% open‑source dependency in AI supply chains. Practical fixes include scanning downloaded models, signing deployments, and building an AI bill of materials (AIBOM) for forensics. Operational risks like model drift monitoring and secrets leaked in Jupyter Notebooks get concrete mitigation steps, red‑team ideas, and integration points with DevSecOps.
AI-written summary · about
chapters (9)
- 0:00Understanding AI security
- 17:38ML lifecycle and operations
- 21:59Implementing MLSecOps
- 27:48Importance of AI bills of materials
- 30:39Managing AI supply chains
- 34:35Risk management frameworks
- 36:55Building internal security
- 39:17Threat modeling and data hygiene
- 43:56Ongoing security practices
related talks
42mDevSecOps for AI: Introducing MLSecOps for Software 2.0
42mMLOps Under Attack: Threat Modeling Modern AI Systems
18mCommit Secure AI: Embedding Security in Code, Pipelines, and Production
26mRunning Secure AI/ML Experiment in a Unified MLOPS-DevSecOps Pipeline
27mA Gentle Introduction to MLSecOps
12mSecuring AI-Driven Finance
35mGenAI Application Security: Not Just Prompt Injection
45mHardening AI/ML Systems - The Next Frontier of Cybersecurity