How LLMs Transform Verification & Validation for Safety-Critical Autonomous Systems

Session details:

Ensuring the reliability and safety of autonomous systems requires rigorous verification and validation (V&V) across complex perception, planning, and control pipelines. Traditional approaches—rule-based checks, simulation-driven test suites, and human-in-the-loop evaluations—struggle to keep pace with the scale, variability, and uncertainty inherent in safety-critical domains. Recent advances in Large Language Models (LLMs) offer new opportunities to reimagine the V&V lifecycle. This session explores how LLMs can act as dynamic test oracles, scenario generators, and explainability engines to accelerate defect discovery, enhance safety compliance, and reduce validation costs. We highlight case studies in autonomous driving and in-car ecosystems, where LLMs are applied to automate requirement-to-test translation, generate adversarial edge cases, and provide interpretable safety incident narratives for engineers and regulators. Finally, we discuss open challenges around trustworthiness, bias, and regulatory acceptance, outlining a roadmap for responsibly integrating LLMs into safety-critical testing pipelines.