Image de couverture de livre

Disentangling Intent from Role: Adversarial Self-Play for Persona-Invariant Safety Alignment

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Li, Jiajia, Wen, Xiaoyu, Ma, Zhongtian, Hu, Shuyue, Zhang, Qiaosheng, Wang, Zhen
Format:	Preprint
Publié:	2026
Sujets:	Artificial Intelligence
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

MAGIC: A Co-Evolving Attacker-Defender Adversarial Game for Robust LLM Safety
par: Wen, Xiaoyu, et autres
Publié: (2026)

TriPlay-RL: Tri-Role Self-Play Reinforcement Learning for LLM Safety Alignment
par: Tan, Zhewen, et autres
Publié: (2026)

Expert Personas Improve LLM Alignment but Damage Accuracy: Bootstrapping Intent-Based Persona Routing with PRISM
par: Hu, Zizhao, et autres
Publié: (2026)

Stay in Character, Stay Safe: Dual-Cycle Adversarial Self-Evolution for Safety Role-Playing Agents
par: Liao, Mingyang, et autres
Publié: (2026)

Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Images
par: Yang, Qishun, et autres
Publié: (2026)

From Persona to Personalization: A Survey on Role-Playing Language Agents
par: Chen, Jiangjie, et autres
Publié: (2024)

Adaptive Theory of Mind for LLM-based Multi-Agent Coordination
par: Mu, Chunjiang, et autres
Publié: (2026)

Memory-Driven Role-Playing: Evaluation and Enhancement of Persona Knowledge Utilization in LLMs
par: Wang, Kai, et autres
Publié: (2026)

Towards Context-Invariant Safety Alignment for Large Language Models
par: Wang, Yixu, et autres
Publié: (2026)

PersonaArena: Dynamic Simulation for Evaluating and Enhancing Persona-Level Role-Playing in Large Language Models
par: Shi, Wenlong, et autres
Publié: (2026)

Self-Play with Adversarial Critic: Provable and Scalable Offline Alignment for Language Models
par: Ji, Xiang, et autres
Publié: (2024)

Enhancing Persona Following at Decoding Time via Dynamic Importance Estimation for Role-Playing Agents
par: Liu, Yuxin, et autres
Publié: (2026)

Graph Attention is Not Always Beneficial: A Theoretical Analysis of Graph Attention Mechanisms via Contextual Stochastic Block Models
par: Ma, Zhongtian, et autres
Publié: (2024)

Seirênes: Adversarial Self-Play with Evolving Distractions for LLM Reasoning
par: Zhang, Chi, et autres
Publié: (2026)

RoleCDE:Benchmarking and Mitigating Role-Alignment Trade-offs in Role-Playing Agents
par: Lai, Huayi, et autres
Publié: (2026)

Character is Destiny: Can Role-Playing Language Agents Make Persona-Driven Decisions?
par: Xu, Rui, et autres
Publié: (2024)

Dynamic Adversarial Reinforcement Learning for Robust Multimodal Large Language Models
par: Bao, Yicheng, et autres
Publié: (2026)

Matrix Completion with Hypergraphs:Sharp Thresholds and Efficient Algorithms
par: Ma, Zhongtian, et autres
Publié: (2024)

DPRF: A Generalizable Dynamic Persona Refinement Framework for Optimizing Behavior Alignment Between Personalized LLM Role-Playing Agents and Humans
par: Yao, Bingsheng, et autres
Publié: (2025)

Persona-Aware Alignment Framework for Personalized Dialogue Generation
par: Li, Guanrong, et autres
Publié: (2025)

LLMRouterBench: A Massive Benchmark and Unified Framework for LLM Routing
par: Li, Hao, et autres
Publié: (2026)

OOD-MMSafe: Advancing MLLM Safety from Harmful Intent to Hidden Consequences
par: Wen, Ming, et autres
Publié: (2026)

Large Language Models are Near-Optimal Decision-Makers with a Non-Human Learning Behavior
par: Li, Hao, et autres
Publié: (2025)

Multilingual Safety Alignment via Self-Distillation
par: Qin, Ruiyang, et autres
Publié: (2026)

CharacterGPT: A Persona Reconstruction Framework for Role-Playing Agents
par: Park, Jeiyoon, et autres
Publié: (2024)

ICL-Router: In-Context Learned Model Representations for LLM Routing
par: Wang, Chenxu, et autres
Publié: (2025)

The Alignment Floor: How Persona Customization Breaks Safety in Weakly-Aligned LLMs
par: Zhang, Xing, et autres
Publié: (2026)

SPC: Evolving Self-Play Critic via Adversarial Games for LLM Reasoning
par: Chen, Jiaqi, et autres
Publié: (2025)

PerPilot: Personalizing VLM-based Mobile Agents via Memory and Exploration
par: Wang, Xin, et autres
Publié: (2025)

Learning Robust Reasoning through Guided Adversarial Self-Play
par: Li, Shuozhe, et autres
Publié: (2026)

Role-Augmented Intent-Driven Generative Search Engine Optimization
par: Chen, Xiaolu, et autres
Publié: (2025)

ELITE: Experiential Learning and Intent-Aware Transfer for Self-improving Embodied Agents
par: Wei, Bingqing, et autres
Publié: (2026)

Stochastic Monkeys at Play: Random Augmentations Cheaply Break LLM Safety Alignment
par: Vega, Jason, et autres
Publié: (2024)

On-Policy Self-Evolution via Failure Trajectories for Agentic Safety Alignment
par: Yin, Bo, et autres
Publié: (2026)

CoSER: A Comprehensive Literary Dataset and Framework for Training and Evaluating LLM Role-Playing and Persona Simulation
par: Wang, Xintao, et autres
Publié: (2025)

Style Attack Disguise: When Fonts Become a Camouflage for Adversarial Intent
par: Zhang, Yangshijie, et autres
Publié: (2025)

Balancing the AI Strength of Roles in Self-Play Training with Regret Matching+
par: Wang, Xiaoxi
Publié: (2024)

Ensemble Successor Representations for Task Generalization in Offline-to-Online Reinforcement Learning
par: Wang, Changhong, et autres
Publié: (2024)

MINDECHO: Role-Playing Language Agents for Key Opinion Leaders
par: Xu, Rui, et autres
Publié: (2024)

Do We Truly Need So Many Samples? Multi-LLM Repeated Sampling Efficiently Scales Test-Time Compute
par: Chen, Jianhao, et autres
Publié: (2025)