ARES: Adaptive Red-Teaming and End-to-End Repair of Policy-Reward System (2026.acl-long)
Copied to clipboard
Jiacheng Liang, Yao Ma, Tharindu Kumarage, Satyapriya Krishna, Rahul Gupta, Kai-Wei Chang, Aram Galstyan, Charith Peris
| Challenge: | Existing red-teaming approaches focus on policy-level weaknesses, but they overlook systemic weaknesses . aRES exploits dual-targeting weaknesses in both the core LLM and the RM simultaneously. |
| Approach: | a new framework uncovers weaknesses in both the core and the reward models simultaneously . a "Safety Mentor" generates semantically coherent adversarial prompts . |
| Outcome: | ARES uncovers weaknesses in both the core LLM and the RM simultaneously . it fine-tunes the LM to detect harmful content, then optimizes the core model . |
Similar Papers
Atoxia: Red-teaming Large Language Models with Target Toxic Answers (2025.findings-naacl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are still vulnerable to generation safety vulnerabilities. |
| Approach: | They propose a method that A**tacks LLMs with target "toxi" given a particular harmful answer, the method generates a user query and a misleading answer opening to examine the internal defects of a given LLM. |
| Outcome: | The proposed method detects safety risks in open-source models and state-of-the-art models such as GPT-4o. |
RLHFPoison: Reward Poisoning Attack for Reinforcement Learning with Human Feedback in Large Language Models (2024.acl-long)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models (LLMs) have significantly enhanced the capabilities in natural language processing. |
| Approach: | They propose a method to poison large language models by using annotators to rank a set of collected responses to generate longer tokens. |
| Outcome: | The proposed method can generate longer tokens without harming the original safety alignment performance. |
Be Your Own Red Teamer: Safety Alignment via Self-Play and Reflective Experience Replay (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have remarkable capabilities but are vulnerable to adversarial “jailbreak” attacks designed to bypass safety guardrails. |
| Approach: | They propose to empower a large language model to be its own red teamer . safety self-play allows the model to act as both the Attacker and Defender . |
| Outcome: | The proposed approach outperforms baselines trained on static adversarial datasets and establishes a new benchmark for proactive safety alignment. |
Automated Progressive Red Teaming (2025.coling-main)
Copied to clipboard
| Challenge: | Automated red teaming (ART) is effective but time-consuming, costly and lacks scalability. |
| Approach: | They propose an automated red teaming framework that generates adversarial prompts to expose LLM vulnerabilities. |
| Outcome: | The proposed framework explores and exploits LLM vulnerabilities through multi-round interactions. |
Holistic Automated Red Teaming for Large Language Models through Top-Down Test Case Generation and Multi-turn Interaction (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches focus on improving attack success rates while overlooking the need for comprehensive test case coverage. |
| Approach: | They propose a top-down approach to automated red teaming that scales up the diversity of test cases using an extensible, fine-grained risk taxonomy. |
| Outcome: | The proposed approach scales up the diversity of test cases using a top-down approach based on an extensible, fine-grained risk taxonomy and leverages reinforcement learning techniques to facilitate multi-turn adversarial probing in a human-like manner. |
RLShield: Dynamic Jailbreak Detection for LLMs via Reinforced Adaptive Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to detect jailbreak prompts rely on static model components or fixed decision thresholds. |
| Approach: | They propose a dynamic jailbreak detection framework that employs reinforcement learning for adaptive threshold selection. |
| Outcome: | Experimental results show that the framework outperforms baselines in detection performance while maintaining high computational efficiency. |
TRIDENT: Enhancing Large Language Model Safety with Tri-Dimensional Diversified Red-Teaming Data Synthesis (2025.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) excel in natural language processing tasks but are vulnerable to harmful content and being exploited for malicious purposes. |
| Approach: | They propose a framework to measure the risk coverage of alignment datasets across three dimensions: Lexical Diversity, Malicious Intent, and Jailbreak Tactics. |
| Outcome: | The proposed framework measures risk coverage across Lexical Diversity, Malicious Intent, and Jailbreak Tactics. |
Adversarial Preference Learning for Robust LLM Alignment (2025.findings-acl)
Copied to clipboard
Yuanfu Wang, Pengyu Wang, Chenyang Xi, Bo Tang, Junyi Zhu, Wenqiang Wei, Chen Chen, Chao Yang, Jingfeng Zhang, Chaochao Lu, Yijun Niu, Keming Mao, Zhiyu Li, Feiyu Xiong, Jie Hu, Mingchuan Yang
| Challenge: | Modern language models rely on Reinforcement Learning from Human Feedback (RLHF) to encourage safe behaviors, but they remain vulnerable to adversarial attacks due to three key limitations: (1) the inefficiency and high cost of human annotation; (2) the vast diversity of potential adversarials; and (3) the risk of feedback bias and reward hacking. |
| Approach: | They propose an iterative adversarial training method that incorporates three key innovations to address these challenges. |
| Outcome: | Experiments on Mistral-7B-Instruct-v0.3 show that the proposed method significantly enhances robustness and reduces harmful outputs from 5.88% to 0.43%. |
TriPlay-RL: Tri-Role Self-Play Reinforcement Learning for LLM Safety Alignment (2026.acl-long)
Copied to clipboard
Zhewen Tan, Wenhan Yu, Jianfeng Si, Tongxin Liu, Kaiqi Guan, Huiyan Jin, Jiawen Tao, Xiaokun Yuan, Xiangzheng Zhang, Duohe Ma, Tong Yang, Lin Sun
| Challenge: | Existing approaches to safety alignment of large language models rely on costly manual annotations or human review. |
| Approach: | They propose a closed-loop reinforcement learning framework called TriPlay-RL that enables iterative collaboration among three roles with near-zero manual annotation. |
| Outcome: | The proposed framework achieves 20%–50% improvement in adversarial effectiveness while preserving high output diversity while achieving 10%–30% gains in safety performance without degrading general reasoning capability. |
RED: Unleashing Token-Level Rewards from Holistic Feedback via Reward Redistribution (2025.emnlp-main)
Copied to clipboard
| Challenge: | Experimental results demonstrate the superiority of our approach to aligning large language models with human preferences. |
| Approach: | They propose a method that evaluates and assigns specific credit to each token using an off-the-shelf reward model. |
| Outcome: | The proposed method evaluates and assigns specific credit to each token using an off-the-shelf reward model. |