Software Security in the Age of AI: Why Traditional SBOMs Are No Longer Enough

In this episode of UC Today, host Kristian McCann is joined by Devashri Datta, a Senior Security Architect at NVIDIA, but appearing in a personal capacity as a researcher, to explore why organizations need to rethink how they govern AI-powered tools.  

Devashri explains why Software Bills of Materials (SBOMs) must evolve to account for AI models, training data, agent behavior, and new attack methods such as prompt injection and data poisoning. 

From Microsoft Copilot and Slack integrations to autonomous AI agents capable of accessing sensitive business systems, this conversation explores the hidden risks enterprises face as AI adoption accelerates. Watch to understand why AI governance needs to move at machine speed and why treating AI like “just another tool” could leave organizations exposed. 

Key topics discussed include: 
🔵 Why traditional SBOMs struggle to capture AI-specific risks, including model weights, training data, RAG pipelines, and AI behavior.  
🔵 How AI agents introduce new security challenges through prompt injection, data poisoning, excessive permissions, and access creep.  
🔵 Why organizations need new governance approaches, including AI-focused vulnerability tracking, threat modeling, and machine-speed security controls.  
🔵 How businesses can build stronger AI security strategies while keeping humans involved in critical decisions.  

Share your thoughts in the comments and subscribe to UC Today for more conversations on cybersecurity, AI, collaboration, and enterprise technology.

For more Unified Communications & Collaboration Tech News visit https://www.uctoday.com/