The MLSecOps Podcast AI Security: Map It, Manage It, Master It
Mar 13, 2025
In this engaging discussion, security veteran Brian Pendleton, a prominent researcher in AI security, delves into the crucial need for mapping AI integrations to identify vulnerabilities. He highlights the advantages of using Software Bill of Materials (SBOMs) for risk management over model cards. Bridging communication between machine learning and security teams is vital, and Brian underscores that AI should be treated like traditional software. His recommendations include cataloging AI assets and tackling shadow AI through effective monitoring. The conversation sets the stage for advanced AI risk management strategies.
AI Snips
Chapters
Transcript
Episode notes
From TRS-80 To AI Village
- Brian Pendleton traces his hacker origins to a childhood TRS-80 and early university network tinkering.
- He joined AI Village in 2017 seeking real-world AI security problems beyond academic adversarial ML research.
AI Security Is A New Field
- Claims of decades of AI security experience are often misleading because formal AI security is recent.
- Real AI security work only became practical around 2017–2018, not 20 years ago.
Hire ML Security Engineers
- Create ML security engineer roles to bridge data science and security teams.
- Use those engineers to translate risks for management and align teams on mitigations.