Commit Secure AI: Embedding Security in Code, Pipelines, and Production
summary
Claims that AI powered attacks are three times more successful and that threat modeling catches ~50% of vulnerabilities drive the agenda: show actual attacks (prompt injection, data poisoning, evasion, model inversion), demonstrate red teaming vs pen testing, and prove runtime protection by logging every token and session to block malicious actors. Concrete tooling includes Microsoft Threat Modeler, MLflow, Prometheus and Grafana, model registries, dependency and container scans, automated responses, and documented model supply‑chain and versioning practices.
AI-written summary · about
chapters (9)
- 0:00Welcome
- 2:10Why AI security matters
- 5:20Changing AI security landscape
- 7:49Security by design in AI
- 11:43Securing AI model supply chain
- 12:47Testing AI models for vulnerabilities
- 13:52Runtime monitoring for AI
- 15:19AI security roadmap and tools
- 17:08Summary
related talks
50mBuilding AI Security In: MLSecOps in Practice
45mThreat Modeling in the Age of AI
25mThreat Modeling for AI Apps with Attacks as Code
35mGenAI Application Security: Not Just Prompt Injection
49mThe Intersection of AI and Cybersecurity
47mInside AI Security with
19mSecuring AI Pipelines from Development to Production
37mFundamentals of AI safety and security