paper-with-me

홈 › Papers

Backdoors in RLVR: Jailbreak Backdoors in LLMs From Verifiable Reward

2026-04-10 · Weiyang Guo, Zesheng Shi, Zeen Zhu, Yuan Zhou, Min Zhang, Jing Li arxiv

Reinforcement Learning with Verifiable Rewards (RLVR) is an emerging paradigm that significantly boosts a Large Language Model's (LLM's) reasoning abilities on complex logical tasks, such as mathematics and programming. However, we identify, for the first time, a latent vulnerability to backdoor attacks within the RLVR framework. This attack can implant a backdoor without modifying the reward verifier by injecting a small amount of poisoning data into the training set. Specifically, we propose a novel trigger mechanism designated as the \ourapproach (ACB). The attack exploits the RLVR training loop by assigning substantial positive rewards for harmful responses and negative rewards for refusals. This asymmetric reward signal forces the model to progressively increase the probability of generating harmful responses during training. Our findings demonstrate that the RLVR backdoor attack is characterized by both high efficiency and strong generalization capabilities. Utilizing less than 2\% poisoned data in train set, the backdoor can be successfully implanted across various model scales without degrading performance on benign tasks. Evaluations across multiple jailbreak benchmarks indicate that activating the trigger degrades safety performance by an average of 73\%. Furthermore, the attack generalizes effectively to a wide range of jailbreak methods and unsafe behaviors. Code is available at https://github.com/yuki-younai/Backdoor_in_RLVR.

📄 PDF Abstract BibTeX arXiv:2604.09748

Code (0)

등록된 구현이 없습니다.

Tasks

Reinforcement Learning

Similar Papers 제목 키워드 기반

Latent Instruction Representation Alignment: defending against jailbreaks, backdoors and undesired knowledge in LLMs

2026-04-12 · Eric Easley, Sebastian Farquhar arxiv

We address jailbreaks, backdoors, and unlearning for large language models (LLMs). Unlike prior work, which trains LLMs based on their actions when given malign instructions, our method specifically trains the model to c…

Universal Jailbreak Backdoors from Poisoned Human Feedback

2023-11-24 · Javier Rando, Florian Tramèr

Reinforcement Learning from Human Feedback (RLHF) is used to align large language models to produce helpful and harmless responses. Yet, prior work showed these models can be jailbroken by finding adversarial prompts tha…

Backdoor Attack

Competition Report: Finding Universal Jailbreak Backdoors in Aligned LLMs

2024-04-22 · Javier Rando, Francesco Croce, Kryštof Mitka, Stepan Shabalin 외

Large language models are aligned to be safe, preventing users from generating harmful content like misinformation or instructions for illegal activities. However, previous work has shown that the alignment process is vu…

Misinformation

Pay Attention to the Triggers: Constructing Backdoors That Survive Distillation

2025-10-21 · Giovanni De Muri, Mark Vero, Robin Staab, Martin Vechev arxiv

LLMs are often used by downstream users as teacher models for knowledge distillation, compressing their capabilities into memory-efficient models. However, as these teacher models may stem from untrusted parties, distill…

Knowledge Distillation

ToxScreen: Detecting Whether an LLM Has Been Poisoned

2026-07-29 · Anthony Hughes, Nicole Xing, Collin Francel, Andy Kim 외 arxiv

As large language models (LLMs) are deployed in high-stakes domains, adversaries may poison training data to implant backdoors: hidden triggers that covertly manipulate model behavior at inference time. We ask whether a …