Webinar: Attacking AI with Fault Injection, Advancing FI with AI Assistance
July 21st, 9AM PST / 6 PM CET
About the webinar:
AI is changing the Fault Injection landscape in two important ways. First, AI-enabled hardware is becoming a target: edge AI systems running on NPUs can be affected by Fault Injection, leading to corrupted inference results such as confidence drift, missed detections, and phantom objects.
Second, AI is becoming part of the testing workflow itself, with LLMs and GPT-assisted approaches showing potential to support FI campaign planning, optimization, and analysis.
In this session, we’ll discuss:
1. Fault Injection on Edge AI Inference
2. Test Setup and Results: YOLOv5 on an NPU Platform
3. Impact on the Object Perception Pipeline
4. Using LLMs/GPTs to Assist Fault Injection Testing
5. Demo: GPT-Assisted FI Workflow in Action
6. Initial Learnings, Conclusion, and Future Work
This webinar is especially relevant for product security teams, embedded AI developers, semiconductor vendors, and organizations evaluating the resilience of edge AI and NPU-based systems.
*The results presented in this webinar are preliminary and part of ongoing research. Further testing and analysis are needed to validate findings across additional platforms, models, and deployment scenarios.*
