Copertina

Mind the Inconspicuous: Revealing the Hidden Weakness in Aligned LLMs' Refusal Boundaries

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Yu, Jiahao, Luo, Haozheng, Hu, Jerry Yao-Chieh, Guo, Wenbo, Liu, Han, Xing, Xinyu
Natura:	Preprint
Pubblicazione:	2024
Soggetti:	Artificial Intelligence
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

Decoupled Alignment for Robust Plug-and-Play Adaptation
di: Luo, Haozheng, et al.
Pubblicazione: (2024)

Fast and Low-Cost Genomic Foundation Models via Outlier Removal
di: Luo, Haozheng, et al.
Pubblicazione: (2025)

Contrastive Reasoning Alignment: Reinforcement Learning from Hidden Representations
di: Luo, Haozheng, et al.
Pubblicazione: (2026)

BlockScan: Detecting Anomalies in Blockchain Transactions
di: Yu, Jiahao, et al.
Pubblicazione: (2024)

Can LLMs Refuse Questions They Do Not Know? Measuring Knowledge-Aware Refusal in Factual Tasks
di: Pan, Wenbo, et al.
Pubblicazione: (2025)

Observer, Not Player: Simulating Theory of Mind in LLMs through Game Observation
di: Wang, Jerry, et al.
Pubblicazione: (2025)

Provably Optimal Memory Capacity for Modern Hopfield Models: Transformer-Compatible Dense Associative Memories as Spherical Codes
di: Hu, Jerry Yao-Chieh, et al.
Pubblicazione: (2024)

Attention Mechanism, Max-Affine Partition, and Universal Approximation
di: Liu, Hude, et al.
Pubblicazione: (2025)

Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training
di: Yuan, Youliang, et al.
Pubblicazione: (2024)

On Statistical Rates and Provably Efficient Criteria of Latent Diffusion Transformers (DiTs)
di: Hu, Jerry Yao-Chieh, et al.
Pubblicazione: (2024)

On Computational Limits of Modern Hopfield Models: A Fine-Grained Complexity Analysis
di: Hu, Jerry Yao-Chieh, et al.
Pubblicazione: (2024)

Why Do Aligned LLMs Remain Jailbreakable: Refusal-Escape Directions, Operator-Level Sources, and Safety-Utility Trade-off
di: Chen, Yu, et al.
Pubblicazione: (2026)

In-Context Algorithm Emulation in Fixed-Weight Transformers
di: Hu, Jerry Yao-Chieh, et al.
Pubblicazione: (2025)

Universal Approximation with Softmax Attention
di: Hu, Jerry Yao-Chieh, et al.
Pubblicazione: (2025)

Uniform Memory Retrieval with Larger Capacity for Modern Hopfield Models
di: Wu, Dennis, et al.
Pubblicazione: (2024)

Transformer Approximations from ReLUs
di: Hu, Jerry Yao-Chieh, et al.
Pubblicazione: (2026)

Minimalist Softmax Attention Provably Learns Constrained Boolean Functions
di: Hu, Jerry Yao-Chieh, et al.
Pubblicazione: (2025)

Conv-CoA: Improving Open-domain Question Answering in Large Language Models via Conversational Chain-of-Action
di: Pan, Zhenyu, et al.
Pubblicazione: (2024)

On Flow Matching KL Divergence
di: Su, Maojiang, et al.
Pubblicazione: (2025)

Synergistic Weak-Strong Collaboration by Aligning Preferences
di: Jiao, Yizhu, et al.
Pubblicazione: (2025)

Learning the Boundary of Solvability: Aligning LLMs to Detect Unsolvable Problems
di: Peng, Dengyun, et al.
Pubblicazione: (2025)

Linearly Decoding Refused Knowledge in Aligned Language Models
di: Shrivastava, Aryan, et al.
Pubblicazione: (2025)

GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Prompts
di: Yu, Jiahao, et al.
Pubblicazione: (2023)

A Survey on Explainable Deep Reinforcement Learning
di: Cheng, Zelei, et al.
Pubblicazione: (2025)

Deep Hidden Cognition Facilitates Reliable Chain-of-Thought Reasoning
di: Chen, Zijun, et al.
Pubblicazione: (2025)

Question Classification with Deep Contextualized Transformer
di: Luo, Haozheng, et al.
Pubblicazione: (2019)

Nonparametric Modern Hopfield Models
di: Hu, Jerry Yao-Chieh, et al.
Pubblicazione: (2024)

LatentRefusal: Latent-Signal Refusal for Unanswerable Text-to-SQL Queries
di: Ren, Xuancheng, et al.
Pubblicazione: (2026)

From Rogue to Safe AI: The Role of Explicit Refusals in Aligning LLMs with International Humanitarian Law
di: Mavi, John, et al.
Pubblicazione: (2025)

Differentially Private Kernel Density Estimation
di: Liu, Erzhi, et al.
Pubblicazione: (2024)

Can Data-Driven Dynamics Reveal Hidden Physics? There Is A Need for Interpretable Neural Operators
di: Gao, Wenhan, et al.
Pubblicazione: (2025)

Learning to Refuse: Towards Mitigating Privacy Risks in LLMs
di: Liu, Zhenhua, et al.
Pubblicazione: (2024)

Building Coding Agents via Entropy-Enhanced Multi-Turn Preference Optimization
di: Yu, Jiahao, et al.
Pubblicazione: (2025)

GPO: Learning from Critical Steps to Improve LLM Reasoning
di: Yu, Jiahao, et al.
Pubblicazione: (2025)

Think Before Refusal : Triggering Safety Reflection in LLMs to Mitigate False Refusal Behavior
di: Si, Shengyun, et al.
Pubblicazione: (2025)

RefusalBench: Why Refusal Rate Misranks Frontier LLMs on Biological Research Prompts
di: Weidener, Lukas, et al.
Pubblicazione: (2026)

Outlier-Efficient Hopfield Layers for Large Transformer-Based Models
di: Hu, Jerry Yao-Chieh, et al.
Pubblicazione: (2024)

LLM Whisperer: An Inconspicuous Attack to Bias LLM Responses
di: Lin, Weiran, et al.
Pubblicazione: (2024)

ICR Probe: Tracking Hidden State Dynamics for Reliable Hallucination Detection in LLMs
di: Zhang, Zhenliang, et al.
Pubblicazione: (2025)

Beyond No: Quantifying AI Over-Refusal and Emotional Attachment Boundaries
di: Noever, David, et al.
Pubblicazione: (2025)