Cover Image

When Refusals Fail: Unstable Safety Mechanisms in Long-Context LLM Agents

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Hadeliya, Tsimur, Jauhar, Mohammad Ali, Sakpal, Nidhi, Cruz, Diogo
Format:	Preprint
Published:	2025
Subjects:	Machine Learning Artificial Intelligence Computation and Language
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Evaluation of Few-Shot Learning for Classification Tasks in the Polish Language
by: Hadeliya, Tsimur, et al.
Published: (2024)

Detection Is Cheap, Routing Is Learned: Why Refusal-Based Alignment Evaluation Fails
by: Frank, Gregory N.
Published: (2026)

LongSafety: Enhance Safety for Long-Context LLMs
by: Huang, Mianqiu, et al.
Published: (2024)

PEEK: Context Map as an Orientation Cache for Long-Context LLM Agents
by: Gu, Zhuohan, et al.
Published: (2026)

When LLM Judge Scores Look Good but Best-of-N Decisions Fail
by: Landesberg, Eddie
Published: (2026)

Why Reasoning Fails to Plan: A Planning-Centric Analysis of Long-Horizon Decision Making in LLM Agents
by: Wang, Zehong, et al.
Published: (2026)

LazyLLM: Dynamic Token Pruning for Efficient Long Context LLM Inference
by: Fu, Qichen, et al.
Published: (2024)

MemAgent: Reshaping Long-Context LLM with Multi-Conv RL-based Memory Agent
by: Yu, Hongli, et al.
Published: (2025)

ResearchAgent: Iterative Research Idea Generation over Scientific Literature with Large Language Models
by: Baek, Jinheon, et al.
Published: (2024)

When Chain-of-Thought Fails, the Solution Hides in the Hidden States
by: Mehrafarin, Houman, et al.
Published: (2026)

RefusalGuard: Geometry-Preserving Fine-Tuning for Safety in LLMs
by: Asif, Sadia, et al.
Published: (2026)

RefusalBench: Generative Evaluation of Selective Refusal in Grounded Language Models
by: Muhamed, Aashiq, et al.
Published: (2025)

When Thoughts Meet Facts: Reusable Reasoning for Long-Context LMs
by: Jeong, Soyeong, et al.
Published: (2025)

AgentFold: Long-Horizon Web Agents with Proactive Context Management
by: Ye, Rui, et al.
Published: (2025)

AcademicEval: Live Long-Context LLM Benchmark
by: Zhang, Haozhen, et al.
Published: (2025)

Too Long, Didn't Model: Decomposing LLM Long-Context Understanding With Novels
by: Hamilton, Sil, et al.
Published: (2025)

InfLLM: Training-Free Long-Context Extrapolation for LLMs with an Efficient Context Memory
by: Xiao, Chaojun, et al.
Published: (2024)

Current Agents Fail to Leverage World Model as Tool for Foresight
by: Qian, Cheng, et al.
Published: (2026)

Enhancing LLM Agent Safety via Causal Influence Prompting
by: Hahm, Dongyoon, et al.
Published: (2025)

LLM Maybe LongLM: Self-Extend LLM Context Window Without Tuning
by: Jin, Hongye, et al.
Published: (2024)

CaRT: Teaching LLM Agents to Know When They Know Enough
by: Liu, Grace, et al.
Published: (2025)

When Stability Fails: Hidden Failure Modes Of LLMS in Data-Constrained Scientific Decision-Making
by: Riasat, Nazia
Published: (2026)

Where Do Reasoning Models Refuse?
by: Yamaguchi, Kureha, et al.
Published: (2025)

Programming Refusal with Conditional Activation Steering
by: Lee, Bruce W., et al.
Published: (2024)

When Safety Blocks Sense: Measuring Semantic Confusion in LLM Refusals
by: Anonto, Riad Ahmed, et al.
Published: (2025)

Beyond Numeric Rewards: In-Context Dueling Bandits with LLM Agents
by: Xia, Fanzeng, et al.
Published: (2024)

TTKV: Temporal-Tiered KV Cache for Long-Context LLM Inference
by: Dzikanyanga, Gradwell, et al.
Published: (2026)

PQCache: Product Quantization-based KVCache for Long Context LLM Inference
by: Zhang, Hailin, et al.
Published: (2024)

When is Tree Search Useful for LLM Planning? It Depends on the Discriminator
by: Chen, Ziru, et al.
Published: (2024)

Evaluating Very Long-Term Conversational Memory of LLM Agents
by: Maharana, Adyasha, et al.
Published: (2024)

Does Refusal Training in LLMs Generalize to the Past Tense?
by: Andriushchenko, Maksym, et al.
Published: (2024)

Refusal in Language Models Is Mediated by a Single Direction
by: Arditi, Andy, et al.
Published: (2024)

A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis
by: Gur, Izzeddin, et al.
Published: (2023)

Federated In-Context LLM Agent Learning
by: Wu, Panlong, et al.
Published: (2024)

100-LongBench: Are de facto Long-Context Benchmarks Literally Evaluating Long-Context Ability?
by: Yang, Wang, et al.
Published: (2025)

LightTransfer: Your Long-Context LLM is Secretly a Hybrid Model with Effortless Adaptation
by: Zhang, Xuan, et al.
Published: (2024)

Tracing the Dynamics of Refusal: Exploiting Latent Refusal Trajectories for Robust Jailbreak Detection
by: Hu, Xulin, et al.
Published: (2026)

Revisiting In-Context Learning with Long Context Language Models
by: Baek, Jinheon, et al.
Published: (2024)

SampleAttention: Near-Lossless Acceleration of Long Context LLM Inference with Adaptive Structured Sparse Attention
by: Zhu, Qianchao, et al.
Published: (2024)

The Geometry of Refusal in Large Language Models: Concept Cones and Representational Independence
by: Wollschläger, Tom, et al.
Published: (2025)