Yingshui Tan

We propose VoTA, a novel attack framework that exploits the tension between logical reasoning and safety objectives in VLMs by generating chains of images with risky visual thoughts, achieving significantly higher success rates than existing methods.

We propose KORGym, a dynamic, game‐based benchmark offering over 50+ interactive tasks with RL support for multi‐turn LLM reasoning evaluation, and validate its effectiveness through extensive experiments, revealing several key insights.