Saved in:
Bibliographic Details
Main Authors: Liu, Yuexiao, Li, Lijun, Wang, Xingjun, Shao, Jing
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.15499
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918162547605504
author Liu, Yuexiao
Li, Lijun
Wang, Xingjun
Shao, Jing
author_facet Liu, Yuexiao
Li, Lijun
Wang, Xingjun
Shao, Jing
contents Recent advancements in Reinforcement Learning with Verifiable Rewards (RLVR) have gained significant attention due to their objective and verifiable reward signals, demonstrating strong performance in reasoning and code generation tasks. However, the potential safety risks associated with RLVR remain underexplored. This paper presents HarmRLVR, the first systematic investigation into the alignment reversibility risk of RLVR. We show that safety alignment can be rapidly reversed using GRPO with merely 64 harmful prompts without responses, causing models to readily comply with harmful instructions. Across five models from Llama, Qwen, and DeepSeek, we empirically demonstrate that RLVR-based attacks elevate the average harmfulness score to 4.94 with an attack success rate of 96.01\%, significantly outperforming harmful fine-tuning while preserving general capabilities. Our findings reveal that RLVR can be efficiently exploited for harmful alignment, posing serious threats to open-source model safety. Please see our code at https://github.com/lyxx2535/HarmRLVR.
format Preprint
id arxiv_https___arxiv_org_abs_2510_15499
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle HarmRLVR: Weaponizing Verifiable Rewards for Harmful LLM Alignment
Liu, Yuexiao
Li, Lijun
Wang, Xingjun
Shao, Jing
Cryptography and Security
Recent advancements in Reinforcement Learning with Verifiable Rewards (RLVR) have gained significant attention due to their objective and verifiable reward signals, demonstrating strong performance in reasoning and code generation tasks. However, the potential safety risks associated with RLVR remain underexplored. This paper presents HarmRLVR, the first systematic investigation into the alignment reversibility risk of RLVR. We show that safety alignment can be rapidly reversed using GRPO with merely 64 harmful prompts without responses, causing models to readily comply with harmful instructions. Across five models from Llama, Qwen, and DeepSeek, we empirically demonstrate that RLVR-based attacks elevate the average harmfulness score to 4.94 with an attack success rate of 96.01\%, significantly outperforming harmful fine-tuning while preserving general capabilities. Our findings reveal that RLVR can be efficiently exploited for harmful alignment, posing serious threats to open-source model safety. Please see our code at https://github.com/lyxx2535/HarmRLVR.
title HarmRLVR: Weaponizing Verifiable Rewards for Harmful LLM Alignment
topic Cryptography and Security
url https://arxiv.org/abs/2510.15499