Adversary-Aware DPO: Enhancing Safety Alignment in Vision Language Models via Adversarial Training (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Adversary-aware DPO (ADPO) is a training framework that explicitly considers adversary. |
| Approach: | a new framework integrates adversarial training into a pre-trained large language model to enhance safety alignment . adversary-aware DPO provides a framework that explicitly considers adversary . |
| Outcome: | a new training framework outperforms baselines in safety alignment and general utility of large language models. |
Similar Papers
Adversarial DPO: Harnessing Harmful Data for Reducing Toxicity with Minimal Impact on Coherence and Evasiveness in Dialogue Agents (2024.findings-naacl)
Copied to clipboard
| Challenge: | Existing toxicity within large language models can negatively impact the user experience, causing performance degradation. |
| Approach: | They propose an adversarial DPO algorithm that improves direct preference optimization (DPO) by incorporating harmful data into the generative model. |
| Outcome: | The proposed training algorithm improves the model’s resilience against harmful conversations while minimizing performance degradation. |
SAFER: Advancing Safety Alignment via Efficient Ex-Ante Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing alignment methods struggle to cover diverse safety scenarios and remain vulnerable to adversarial attacks. |
| Approach: | They propose a framework for 'S**afety' alignment via e**F**ficient' E**x-Ante-R**easoning that instantiates structured Ex-Ance reasoning and embeds predefined safety rules. |
| Outcome: | The proposed framework enhances safety performance while maintaining usefulness and efficiency. |
On the Vulnerability of Safety Alignment in Open-Access LLMs (2024.findings-acl)
Copied to clipboard
Jingwei Yi, Rui Ye, Qisi Chen, Bin Zhu, Siheng Chen, Defu Lian, Guangzhong Sun, Xing Xie, Fangzhao Wu
| Challenge: | Large language models (LLMs) are susceptible to malicious exploitation, but are often rejected and limited harmfulness is limited. |
| Approach: | They propose two types of reverse alignment techniques: reverse supervised fine-tuning (RSFT) and reverse preference optimization (RPO). |
| Outcome: | The proposed methods can significantly enhance the success rate and harmfulness of jailbreak attacks, but they face high rejection rates and limited harmfulness. |
Goal-Conditioned DPO: Prioritizing Safety in Misaligned Instructions (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing defense methods focus on aligning the model’s output towards less harmful responses through post-processing or input perturbation. |
| Approach: | They propose a goal-conditioned direct preference optimization technique which is trained to prioritize the system prompt over the user prompt through goal-conditioning and reduces the average Attack Success Rate (ASR) on a wide variety of jailbreak attacks. |
| Outcome: | The proposed approach reduces the average Attack Success Rate (ASR) on a wide variety of jailbreak attacks while maintaining general performance. |
The Art of (Mis)alignment: How Fine-Tuning Methods Effectively Misalign and Realign LLMs in Post-Training (2026.findings-acl)
Copied to clipboard
Rui Zhang, Hongwei Li, Yun Shen, Xinyue Shen, Wenbo Jiang, Guowen Xu, Yang Liu, Michael Backes, Yang Zhang
| Challenge: | Misaligned large language models can magnify harm by exploiting them to undermine safety . et al., 2022b; Bai e.t., 2023): misalignment, realignment and model-specific resistance are important . |
| Approach: | They evaluate four methods to identify a mechanism asymmetry between attack and defense . they find that ORPO is most effective for misalignment, but DPO excels in realignment . |
| Outcome: | The proposed methods show a mechanism asymmetry between attack and defense . the proposed methods excel in realignment, but at the expense of model utility . |
Safety Is Not Universal: The Selective Safety Trap in LLM Alignment (2026.findings-acl)
Copied to clipboard
Iago Alves Brito, Walcy Rios, Julia Soares Dollis, Diogo Fernandes Costa Silva, Arlindo Rodrigues Galvão Filho
| Challenge: | Existing safety evaluations of large language models aggregate harms under generic categories such as "Identity Hate" a bilingual benchmark identifies a selective safety trap, where defense rates vary by up to 42% within the same model solely based on the target group. |
| Approach: | They propose a bilingual adversarial benchmark to audit selective safety in large language models . defense rates vary by up to 42% within the same model solely based on target group . |
| Outcome: | The proposed benchmark identifies a selective safety trap in large language models . defense rates vary by up to 42% within the same model solely based on the target group. |
Jailbreaks as Inference-Time Alignment: A Framework for Understanding Safety Failures in LLMs (2026.eacl-long)
Copied to clipboard
| Challenge: | Large language models are safety-aligned to prevent harmful response generation . prior work on jailbreak effectiveness has focused on analyzing success rate of jailbreaks . |
| Approach: | They propose to frame jailbreaks as inference-time alignment and draw suboptimal bounds . they also propose a Safety-Net to measure how vulnerable an LLM is to jailbreak attacks . |
| Outcome: | a new framework allows researchers to show how vulnerable an LLM is to jailbreaks . a Safety-Net measures how vulnerable the model is to attacks, the authors say . |
Teaching Your Models to Understand Code via Focal Preference Alignment (2025.emnlp-main)
Copied to clipboard
Jie Wu, Haoling Li, Xin Zhang, Xiao Liu, Yangyu Huang, Jianwen Luo, Yizhen Zhang, Zuchao Li, Ruihang Chu, Yujiu Yang, Scarlett Li
| Challenge: | Existing methods for supervised fine-tuning focus on unit test feedback to construct preference pairs. |
| Approach: | They propose a preference alignment framework that mimics human iterative debugging to refine Code LLMs. |
| Outcome: | Experiments show that Preference Learning improves on BigCodeBench and BigCodeBind tasks. |
Defending Against Alignment-Breaking Attacks via Robustly Aligned LLM (2024.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have made significant advancements but can be misused to generate harmful content. |
| Approach: | They propose a Robustly Aligned LLM to defend against alignment-breaking attacks by retraining existing LLMs and using adversarial or handcrafted jailbreaking prompts. |
| Outcome: | The proposed model reduces attack success rates from nearly 100% to around 10% or less. |
SafeInt: Shielding Large Language Models from Jailbreak Attacks via Safety-Aware Representation Intervention (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Jailbreak attacks exploit vulnerabilities in large language models to induce undesirable behavior . existing defenses cannot dynamically adjust representations based on harmfulness of queries . |
| Approach: | They propose a representation-aware representation method that shields LLMs from jailbreak attacks . SafeInt relocates jailbreak-related representations into the rejection region . |
| Outcome: | The proposed method outperforms baseline defenses while maintaining utility . it relocates jailbreak-related representations into the rejection region . |