Two local models propose and criticize scenarios and refine them from simulations. The approach aims at traceable test evidence on a laptop.
A team describes an agentic framework executed locally to look for rare situations that could put an autonomous system in difficulty. The requirements, the description of the system and its environment feed into the generation; a second model criticizes each scenario.
Simulation tests produce measurements that direct further exploration towards close or new cases. The grouping of results helps to identify faulty functions and their triggering conditions. The work presents a method of producing elements for safety assurance, without claiming to cover all possible risks.
Preprint filed on arXiv on October 9, 2026; results are those reported by the authors and remain to be independently confirmed.
Read the full article on the original source
Go to the source →