Automotive Grade Linux All Member Meeting Japan 2026

While the Software-Defined Vehicle (SDV) era promises rapid updates and new capabilities, the gap between architectural complexity and safety proof is widening faster than traditional audits can manage.
In this talk, the author shares real-world strategies for building a Multi-Party, LLM-Augmented Fuzzing Framework purpose-built for automotive safety and security. This approach demonstrates how to utilize Large Language Models, trained on protocol specifications, to generate semantics-aware seeds. It also explores how feeding Fault Tree Analysis (FTA) into this loop allows teams to exercise previously untested leaf events.
The session concludes by expanding beyond AI fuzzing to the deployment of general AI agents within the vehicle. It highlights the architectural shift toward AI-Native SDVs, covering the necessity of an Agentic runtime and a Policy Guard layer that safely reduces the cost of deploying in-vehicle AI Agents.
The audience will leave the session with a holistic impression of how to practically harness AI—both as a rigorous verification instrument and as a secure, in-vehicle diagnostic agent—to build safer, OSS-based automotive systems.