Search papers, labs, and topics across Lattice.
This paper introduces SafeCap, a reinforcement-learning framework designed to enhance the safety of large vision-language models (LVLMs) by leveraging learned self-captioning. By training a policy model to generate safety-relevant image captions that guide a frozen language model towards safety-aligned decisions, SafeCap significantly improves safety performance across multiple benchmarks. The results show gains of 3.7-19.0 points in safety averages while maintaining or enhancing vision utility, outperforming existing safety alignment methods like safety SFT and DPO.
SafeCap boosts LVLM safety by up to 19 points through innovative caption-mediated reinforcement learning, outpacing traditional alignment methods.
Large vision-language models (LVLMs) remain vulnerable to jailbreak attacks that exploit visual inputs to bypass safety alignment inherited from their language backbones. We propose SafeCap, a reinforcement-learning framework that aligns LVLMs through learned self-captioning. SafeCap trains a policy model to first generate a safety-relevant image caption and then produce a final answer; the caption is further optimized by whether it enables a frozen LLM to reach a safety-aligned decision. This caption-mediated objective encourages the policy to expose visual cues relevant to safe response generation rather than relying solely on direct refusal supervision. Across five multimodal safety benchmarks and six vision-utility benchmarks, SafeCap substantially improves aggregate safety performance under its intended DirectCap protocol, with gains of 3.7-19.0 points in safety average across four model settings while maintaining comparable or improved vision utility. Under controlled comparisons on matched backbones and data, SafeCap outperforms safety SFT, DPO, and SafeGRPO, demonstrating the effectiveness of caption-mediated reinforcement learning for multimodal safety alignment.
