RUBAS: Rubric-Based Reinforcement Learning for Agent Safety

In three linesRUBAS is a rubric-based reinforcement learning framework for LLM agent alignment. It decomposes behavior into four dimensions (tool-use safety, argument safety, response safety, helpfulness) and generates structured rewards over complete trajectories. Experiments show improved safety and reduced hallucinations while maintaining utility.

Summary generated by Claude — human-verified