arXiv:2603.08234v1 Announce Type: new Abstract: With the rapid advancement of large language models (LLMs), the safety of LLMs has become a critical concern. Despite significant efforts in safety alignment, current LLMs remain vulnerable to jailbreaking attacks. However, the root causes of such vulnerabilities are still poorly understood, necessitating a rigorous investigation into jailbreak mechanisms across both academic and industrial communities. In this work, we focus on a continuation-triggered jailbreak phenomenon, whereby simply relocating a continuation-triggered instruction suffix can substantially increase jailbreak success rates. To uncover the intrinsic mechanisms of this phenomenon, we conduct a comprehensive mechanistic interpretability analysis at the level of attention heads. Through causal interventions and activation scaling, we show that this jailbreak behavior primarily arises from an inherent competition between the model's intrinsic continuation drive and the safety defenses acquired through alignment training. Furthermore, we perform a detailed behavioral analysis of the identified safety-critical attention heads, revealing notable differences in the functions and behaviors of safety heads across different model architectures. These findings provide a novel mechanistic perspective for understanding and interpreting jailbreak behaviors in LLMs, offering both theoretical insights and practical implications for improving model safety.

arXiv:2602.04265v2 Announce Type: replace-cross Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a promising paradigm for enhancing reasoning in Large Language Models (LLMs). However, it frequently encounters challenges such as entropy collapse, excessive verbosity, and insufficient exploration for hard problems. Crucially, existing reward schemes fail to distinguish between the need for extensive search during problem-solving and the efficiency required for mastered knowledge. In this work, we introduce T2T(Thickening-to-Thinning), a dynamic reward framework inspired by human learning processes. Specifically, it implements a dual-phase mechanism: (1) On incorrect attempts, T2T incentivizes "thickening" (longer trajectories) to broaden the search space and explore novel solution paths; (2) Upon achieving correctness, it shifts to "thinning", imposing length penalties to discourage redundancy, thereby fostering model confidence and crystallizing reasoning capabilities. Extensive experiments on mathematical benchmarks (MATH-500, AIME, AMC) across Qwen-series and Deepseek models demonstrate that T2T significantly outperforms standard GRPO and recent baselines, achieving superior performance.