Papers by Yaodong Yang
SafeMCP: Proactive Power Regulation for LLM Agent Defense via Environment-Grounded Look-Ahead Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Model (LLM) agents are expanding their action spaces to operate in complex environments. |
| Approach: | They propose a server-side defense plugin that constrains tool acquisition via predictive reasoning regarding future safety risks. |
| Outcome: | Experiments on PowerSeeking Bench, ToolEmu, and AgentHarm show that SafeMCP achieves a safe equilibrium, effectively mitigating risks while preserving agent utility. |
SafeMT: Multi-turn Safety for Multimodal Language Models (2026.acl-long)
Copied to clipboard
Han Zhu, Juntao Dai, Jiaming Ji, Haoran Li, Chengkun Cai, Pengcheng Wen, Chi-Min Chan, Boyuan Chen, Yaodong Yang, Sirui Han, Yike Guo
| Challenge: | Multi-turn dialogues pose a greater risk than single prompts, but existing safety benchmarks do not account for this situation. |
| Approach: | They propose a benchmark that features dialogues of varying lengths generated from harmful queries accompanied by images. |
| Outcome: | The proposed model reduces multi-turn Attack Success Rate (ASR) compared to existing guard models. |
Uncovering Strategic Egoism Behaviors in Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Extensive experiments on 9 proprietary LLMs reveal that SE behaviors are widespread . study identifies egoistic decision-making as a risk for large language models . |
| Approach: | They propose a benchmark to measure egoistic behavior in large language models . they propose toxicity, jailbreak vulnerability and a lightweight mitigation that reinforces situational constraints . |
| Outcome: | The proposed model has a 67.96% occurrence rate and frequently manifests as manipulative coercion. |
Boosting Policy and Process Reward Models with Monte Carlo Tree Search in Open-Domain QA (2025.findings-acl)
Copied to clipboard
Chi-Min Chan, Chunpu Xu, Junqi Zhu, Jiaming Ji, Donghai Hong, Pengcheng Wen, Chunyang Jiang, Zhen Ye, Yaodong Yang, Wei Xue, Sirui Han, Yike Guo
| Challenge: | Experimental results show that our approach can effectively improve the performance of both the policy model and the reward model. |
| Approach: | They propose to use Monte Carlo Tree Search for both policy model improvement and reward model improvement to bridge it to more subtle open-domain question answering. |
| Outcome: | The proposed approach surpasses existing methods for annotation and training data with fewer data points and achieves better performance in test-time scaling strategies. |
A Game-Theoretica Negotiation Framework for Cross-Cultural Consensus (2026.acl-long)
Copied to clipboard
| Challenge: | Large language models exhibit pronounced WEIRD cultural bias, marginalizing diverse viewpoints and posing challenges for reconciling diverse populations with varying cultural backgrounds and value systems. |
| Approach: | They propose a framework for cross-cultural fairness using a Nash Equilibrium . they propose equilibriums that iteratively propose and refine natural-language guidelines . |
| Outcome: | The proposed framework generates higher-quality and more balanced consensus . it finetunes diverse LLM architectures with negotiation data, reducing cultural distances by 95.53%. |
Communication-Efficient Desire Alignment for Proactive Embodied Human–Agent Interaction (2026.acl-long)
Copied to clipboard
| Challenge: | Effective real-world human–agent interactions are long-term and repeated. |
| Approach: | They propose a simulation that uses a proxy user with value-driven preferences and natural language behavior to evaluate how agents adapt to users across interactions and satisfy their desires. |
| Outcome: | HA-Desire, a home assistance simulation, shows that agents can adapt to user needs and provide proactive assistance within limited communication. |
SIV-Bench: A Video Benchmark for Social Interaction Understanding and Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing benchmarks for Multimodal Large Language Models (MLLMs) have been lacking due to the rich nature of social interaction. |
| Approach: | They propose a video benchmark to evaluate MLLMs' capabilities across social scene understanding, social state reasoning, and social dynamics prediction. |
| Outcome: | The proposed benchmarks evaluate MLLMs' capabilities across social scene understanding, social state reasoning, and social dynamics prediction tasks. |
PKU-SafeRLHF: Towards Multi-Level Safety Alignment for LLMs with Human Preference (2025.acl-long)
Copied to clipboard
Jiaming Ji, Donghai Hong, Borong Zhang, Boyuan Chen, Josef Dai, Boren Zheng, Tianyi Alex Qiu, Jiayi Zhou, Kaile Wang, Boxun Li, Sirui Han, Yike Guo, Yaodong Yang
| Challenge: | Using large-scale annotation data, large language models can generate noise, errors and biases, leading to unexpected behaviours. |
| Approach: | They propose a dataset to promote safety alignment in large language models . they separate helpfulness and harmlessness annotations for question-answering pairs . |
| Outcome: | The proposed dataset provides 44.6k prompts and 265k question-answer pairs with safety meta-labels for 19 harm categories and three severity levels, with answers generated by Llama-family models. |
SafeLawBench: Towards Safe Alignment of Large Language Models (2025.findings-acl)
Copied to clipboard
Chuxue Cao, Han Zhu, Jiaming Ji, Qichao Sun, Zhenghao Zhu, Wu Yinyu, Josef Dai, Yaodong Yang, Sirui Han, Yike Guo
| Challenge: | Recent studies indicate that large language models (LLMs) may exhibit risks, including threats to the protection of private data and the generation of hallucinations. |
| Approach: | They propose to evaluate LLMs from a legal perspective using the SafeLawBench benchmark. |
| Outcome: | The proposed framework categorizes safety risks into three levels based on legal standards and includes 24,860 multi-choice questions and 1,106 open-domain question-answering tasks. |
When Slower Isn’t Truer: Inverse Scaling Law of Truthfulness in Multimodal Reasoning (2026.findings-acl)
Copied to clipboard
Sitong Fang, Wenjing Cao, Jiahao Li, Xuyao Wang, Chi-Min Chan, Sirui Han, Juntao Dai, Yike Guo, Yaodong Yang, Jiaming Ji
| Challenge: | a study of slow reasoning models for multimodal reasoning finds that they are more prone to fabricating plausible yet false details when confronted with incomplete or misleading visual inputs. |
| Approach: | They conduct the first systematic study of the inverse scaling law in slow-thinking paradigms for multimodal reasoning. |
| Outcome: | The findings suggest that slower reasoning models are more prone to fabricating false details . the study analyzed 5,000-sample hierarchical prompt dataset by 50 participants . |
Enhancing LLM-Based Social Bot via an Adversarial Learning Framework (2025.emnlp-main)
Copied to clipboard
| Challenge: | Social media platforms provide an ideal testbed for large language models that exhibit human-like behavior. |
| Approach: | They propose an LLM-based social **Bot that enhances human-like generative capabilities through an adversarial learning framework. |
| Outcome: | The proposed framework generates human-like content aligned with diverse user profiles . it exhibits strong social responsiveness, more accurately modeling opinion dynamics . |
Language Models Resist Alignment: Evidence From Data Compression (2025.acl-long)
Copied to clipboard
Jiaming Ji, Kaile Wang, Tianyi Alex Qiu, Boyuan Chen, Jiayi Zhou, Changye Li, Hantao Lou, Josef Dai, Yunhuai Liu, Yaodong Yang
| Challenge: | Large language models (LLMs) may exhibit undesirable behaviors due to the inevitable biases and harmful content present in training. |
| Approach: | They propose to investigate the elasticity of large language models by examining their performance. |
| Outcome: | The proposed model performance declines rapidly before reverting to the pre-training distribution, the authors show . the proposed model weight and code are available at pku-lm-res ist-alignment.github.io. |
Benchmarking Multi-National Value Alignment for Large Language Models (2025.findings-acl)
Copied to clipboard
Chengyi Ju, Weijie Shi, Chengzhong Liu, Jiaming Ji, Jipeng Zhang, Ruiyuan Zhang, Jiajie Xu, Yaodong Yang, Sirui Han, Yike Guo
| Challenge: | Existing studies on large language models focus on ethical reviews, failing to capture the diversity of national values. |
| Approach: | They propose a national value extraction pipeline to efficiently construct value assessment datasets and a model-based model with instruction tagging to process raw data sources. |
| Outcome: | The proposed benchmark evaluates the alignment of LLMs with the values of five major nations: China, the United States, the UK, France, and Germany. |
Reward Generalization in RLHF: A Topological Perspective (2025.findings-acl)
Copied to clipboard
Tianyi Alex Qiu, Fanzhi Zeng, Jiaming Ji, Dong Yan, Kaile Wang, Jiayi Zhou, Yang Han, Josef Dai, Xuehai Pan, Yaodong Yang
| Challenge: | Existing alignment methods share a common topology of information flow, but their alternatives have not been thoroughly explored. |
| Approach: | They propose a theory of reward generalization in reinforcement learning from human feedback . they propose induced Bayesian networks to model the impact of dataset topologies on reward generalisation . |
| Outcome: | The proposed method achieves an average win rate of 65% on three NLP tasks. |