Cloud Security Podcast by Google
EP226 AI Supply Chain Security: Old Lessons, New Poisons, and Agentic Dreams
May 19, 2025
Christine Sizemore, a Cloud Security Architect at Google Cloud, dives into the complexities of AI supply chain security. She highlights the stark differences between AI and traditional software supply chains, using engaging examples like the Suez Canal incident. The discussion uncovers unexpected threats, such as data poisoning, and emphasizes the need for continuous security integration. Sizemore explores organizational pitfalls to avoid and humorously questions whether AI can secure itself—she even shares practical strategies for safeguarding AI systems.
24:39
Episode guests
AI Summary
AI Chapters
Episode notes
Podcast summary created with Snipd AI
Quick takeaways
- Securing the AI software supply chain requires trusted sources for datasets and libraries to prevent vulnerabilities and data poisoning.
- Organizational pitfalls such as silos between security teams and AI developers must be avoided to ensure effective security integration.
Deep dives
Understanding the AI Supply Chain
AI supply chains are highlighted as being particularly fragile and vulnerable, similar to traditional supply chains. A notable example discussed is the Suez Canal blockage, which demonstrated how disruptions can lead to global shortages and highlight the critical need for components to work harmoniously. AI supply chains involve specific elements such as datasets used for training models and the software libraries necessary for that training, like TensorFlow and PyTorch. Ensuring these elements come from trusted sources and are free from vulnerabilities is essential for maintaining the integrity of AI systems.
