Breaking the Safety-Capability Tradeoff: Reinforcement Learning with Verifiable Rewards Maintains Safety Guardrails in LLMs

• URL: http://arxiv.org/abs/2511.21050v1
• Date: Wed, 26 Nov 2025 04:36:34 GMT
• Title: Breaking the Safety-Capability Tradeoff: Reinforcement Learning with Verifiable Rewards Maintains Safety Guardrails in LLMs
• Authors: Dongkyu Derek Cho, Huan Song, Arijit Ghosh Chowdhury, Haotian An, Yawei Wang, Rohit Thekkanal, Negin Sokhandan, Sharlina Keshava, Hannah Marlowe,
• Abstract summary: reinforcement learning with verifiable rewards (RLVR) has emerged as a promising alternative that optimize models on objectively measurable tasks.<n>We present the first comprehensive theoretical and empirical analysis of safety properties in RLVR.<n> Empirically, we conduct extensive experiments across five adversarial safety benchmarks, demonstrating that RLVR can simultaneously enhance reasoning capabilities while maintaining or improving safety guardrails.
• Score: 3.198812241868092
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Fine-tuning large language models (LLMs) for downstream tasks typically exhibit a fundamental safety-capability tradeoff, where improving task performance degrades safety alignment even on benign datasets. This degradation persists across standard approaches including supervised finetuning (SFT) and reinforcement learning from human feedback (RLHF). While reinforcement learning with verifiable rewards (RLVR) has emerged as a promising alternative that optimizes models on objectively measurable tasks, its safety implications remain unexplored. We present the first comprehensive theoretical and empirical analysis of safety properties in RLVR. Theoretically, we derive upper bounds on safety drift under KL-constrained optimization and prove conditions under which safety degradation is eliminated. Empirically, we conduct extensive experiments across five adversarial safety benchmarks, demonstrating that RLVR can simultaneously enhance reasoning capabilities while maintaining or improving safety guardrails. Our comprehensive ablation studies examine the effects of optimization algorithms, model scale, and task domains. Our findings challenge the prevailing assumption of an inevitable safety capability trade-off, and establish that a specific training methodology can achieve both objectives simultaneously, providing insights for the safe deployment of reasoning-capable LLMs.

Related papers

• Safe Reinforcement Learning via Recovery-based Shielding with Gaussian Process Dynamics Models [57.006252510102506]
  Reinforcement learning (RL) is a powerful framework for optimal decision-making and control but often lacks provable guarantees for safety-critical applications.<n>We introduce a novel recovery-based shielding framework that enables safe RL with a provable safety lower bound for unknown and non-linear continuous dynamical systems.
  arXiv  Detail & Related papers  (2026-02-12T22:03:35Z)
• Think-Reflect-Revise: A Policy-Guided Reflective Framework for Safety Alignment in Large Vision Language Models [58.17589701432514]
  Think-Reflect-Revise (TRR) is a training framework designed to enhance the safety alignment of Large Vision Language Models (LVLMs)<n>We first build a Reflective Safety Reasoning (ReSafe) dataset with 5,000 examples that follow a think-reflect-revise process.<n>We then fine-tune the target model using the ReSafe dataset to initialize reflective behavior, and finally reinforce policy-guided reflection through reinforcement learning.
  arXiv  Detail & Related papers  (2025-12-08T03:46:03Z)
• Decoupling Safety into Orthogonal Subspace: Cost-Efficient and Performance-Preserving Alignment for Large Language Models [39.71740443729507]
  We show that LoRA-based Refusal-training enables performance-preserving safety alignment even when trained solely on safety data.<n>We provide both theoretical and experimental evidence that LoRA effectively decouples safety into a low-rank subspace largely to the model's intrinsic transformation space.
  arXiv  Detail & Related papers  (2025-10-10T05:01:44Z)
• UpSafe$^\circ$C: Upcycling for Controllable Safety in Large Language Models [67.91151588917396]
  Large Language Models (LLMs) have achieved remarkable progress across a wide range of tasks, but remain vulnerable to safety risks such as harmful content generation and jailbreak attacks.<n>We propose UpSafe$circ$C, a unified framework for enhancing LLM safety through safety-aware upcycling.<n>Our results highlight a new direction for LLM safety: moving from static alignment toward dynamic, modular, and inference-aware control.
  arXiv  Detail & Related papers  (2025-10-02T16:43:33Z)
• Anchoring Refusal Direction: Mitigating Safety Risks in Tuning via Projection Constraint [52.878820730054365]
  Instruction Fine-Tuning (IFT) has been widely adopted as an effective post-training strategy to enhance abilities of Large Language Models (LLMs)<n>Recent research into the internal mechanisms of LLMs has identified the refusal direction (r-direction) in the hidden states, which plays a pivotal role in governing refusal behavior.<n>To mitigate such drift, our proposed ProCon method introduces a projection-constrained loss term that regularizes the projection magnitude of each training sample's hidden state onto the r-direction.
  arXiv  Detail & Related papers  (2025-09-08T15:24:33Z)
• Rethinking Safety in LLM Fine-tuning: An Optimization Perspective [56.31306558218838]
  We show that poor optimization choices, rather than inherent trade-offs, often cause safety problems, measured as harmful responses to adversarial prompts.<n>We propose a simple exponential moving average (EMA) momentum technique in parameter space that preserves safety performance.<n>Our experiments on the Llama families across multiple datasets demonstrate that safety problems can largely be avoided without specialized interventions.
  arXiv  Detail & Related papers  (2025-08-17T23:46:36Z)
• Saffron-1: Safety Inference Scaling [69.61130284742353]
  SAFFRON is a novel inference scaling paradigm tailored explicitly for safety assurance.<n>Central to our approach is the introduction of a multifurcation reward model (MRM) that significantly reduces the required number of reward model evaluations.<n>We publicly release our trained multifurcation reward model (Saffron-1) and the accompanying token-level safety reward dataset (Safety4M)
  arXiv  Detail & Related papers  (2025-06-06T18:05:45Z)
• Safety Modulation: Enhancing Safety in Reinforcement Learning through Cost-Modulated Rewards [23.15178050525514]
  Safe Reinforcement Learning (Safe RL) aims to train an RL agent to maximize its performance in real-world environments while adhering to safety constraints.<n>We propose a novel safe RL approach called Safety Modulated Policy Optimization (SMPO), which enables safe policy function learning.
  arXiv  Detail & Related papers  (2025-04-03T21:35:22Z)
• How Does Vision-Language Adaptation Impact the Safety of Vision Language Models? [27.46416187893547]
  Vision-Language adaptation (VL adaptation) transforms Large Language Models (LLMs) into Large Vision-Language Models (LVLMs) Despite potential harmfulness due to weakened safety measures, in-depth analysis on the effects of VL adaptation on safety remains under-explored.
  arXiv  Detail & Related papers  (2024-10-10T03:12:03Z)
• Feasibility Consistent Representation Learning for Safe Reinforcement Learning [25.258227763316228]
  We introduce a novel framework named Feasibility Consistent Safe Reinforcement Learning (FCSRL) This framework combines representation learning with feasibility-oriented objectives to identify and extract safety-related information from the raw state for safe RL. Our method is capable of learning a better safety-aware embedding and achieving superior performance than previous representation learning baselines.
  arXiv  Detail & Related papers  (2024-05-20T01:37:21Z)
• Evaluating Model-free Reinforcement Learning toward Safety-critical Tasks [70.76757529955577]
  This paper revisits prior work in this scope from the perspective of state-wise safe RL. We propose Unrolling Safety Layer (USL), a joint method that combines safety optimization and safety projection. To facilitate further research in this area, we reproduce related algorithms in a unified pipeline and incorporate them into SafeRL-Kit.
  arXiv  Detail & Related papers  (2022-12-12T06:30:17Z)
• Conservative Safety Critics for Exploration [120.73241848565449]
  We study the problem of safe exploration in reinforcement learning (RL) We learn a conservative safety estimate of environment states through a critic. We show that the proposed approach can achieve competitive task performance while incurring significantly lower catastrophic failure rates.
  arXiv  Detail & Related papers  (2020-10-27T17:54:25Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.