ML Model Security - Is Your AI Protected?

Welcome to "MLSecOps Connect: Ask the Experts," an educational live stream series from the MLSecOps Community where attendees have the opportunity to hear their own questions answered by a variety of insightful guest speakers.

This is a recording of the session we held on August 28, 2024 with the Chief Architect at Protect AI, Sean Morgan. In prior roles, Sean led production AI/ML deployments in the semiconductor industry, evaluated adversarial machine learning defenses for DARPA research programs, and most recently scaled customers on interactive machine learning solutions at Amazon Web Services (AWS). In his free time, Sean is an active open-source contributor and maintainer, and is the special interest group lead for TensorFlow Addons.

During this MLSecOps Connect session, Sean fielded questions from the community related to security for AI & machine learning (ML), including the importance of ML model scanning and how to get started with scanning models.

Explore with us:

1. What are some solutions for model scanning?
2. How can enterprises protect their AI models from insider threat?
3. Which three design principles are the most important for secure AI infrastructures?
4. What are some of the security threats we can't use automated scans for?
5. Which resources does Sean use and recommend to advance personal knowledge in the context of AI security?
6. How can someone design a zero trust approach for the current large language model (LLM) architectures?
7. What are some recommendations for aligning best [AI security] practices and standards in a palatable, won't-slow-down-the-business way?
8. Does Protect AI provide both Red Teaming and Guardrails for LLMs?
9. What types of attacks does Protect AI's model scanning tool cover? Are the attacks domain specific (e.g., attacks on text vs image) or generic? Once the model vulnerabilities are detected, what defenses are available?
10. How can we follow a shift left approach for model security? And more!