AI Systems Are Software Systems
summary
Argues AI systems are ordinary software with familiar vulnerability management: use SBOM and track models as libraries, adopt the NIST SSDF AI profile, and follow KEV patching and CISA 2602 decommission rules. Cites a real CVE (CVE-2024-3660) where a model layer executed arbitrary downloads and insecure defaults got fixed. Notes toolable identifiers like package URLs, tooling gaps, and that Deutsche Bahn processes 500,000 SBOMs/day as proof that scale is solvable.
AI-written summary · about
chapters (8)
- 0:00AI cybersecurity focus
- 2:11AI systems vulnerabilities
- 6:19Secure by design principles
- 13:24Asset management and SBOMs
- 17:38AI engineers in cybersecurity
- 28:42CVE rules for AI vulnerabilities
- 37:55Bridging AI and cybersecurity
- 42:41Q&A
related talks
52mAI Is Just Software, What Could Possibly Go Wrong?
29mIdentifying and Assigning AI Model Vulnerabilities
33mAI in the Security Stack
35mModels and Systems: How to Think About Vulnerabilities and Artificial Intelligence
57mAI Applications Security Puzzle
21mNavigating the Intersection: AI’s Role in Shaping the Secure Open Source Software Ecos...
31mSecuring the Future: Navigating AI Vulnerabilities and Evolving Security Practices
14mDesigning the open and safe AI future