Cover Image

Finding and Reactivating Post-Trained LLMs' Hidden Safety Mechanisms

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Li, Mingjie, Si, Wai Man, Backes, Michael, Zhang, Yang, Wang, Yisen
Format:	Preprint
Published:	2026
Subjects:	Computation and Language Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Are Smarter LLMs Safer? Exploring Safety-Reasoning Trade-offs in Prompting and Fine-Tuning
by: Li, Ang, et al.
Published: (2025)

A Systematic Study of Training-Free Methods for Trustworthy Large Language Models
by: Si, Wai Man, et al.
Published: (2026)

Pruning Unsafe Tickets: A Resource-Efficient Framework for Safer and More Robust LLMs
by: Si, Wai Man, et al.
Published: (2026)

TrustLDM: Benchmarking Trustworthiness in Language Diffusion Models
by: Mo, Yichuan, et al.
Published: (2026)

Efficient Switchable Safety Control in LLMs via Magic-Token-Guided Co-Training
by: Si, Jianfeng, et al.
Published: (2025)

SaLoRA: Safety-Alignment Preserved Low-Rank Adaptation
by: Li, Mingjie, et al.
Published: (2025)

Hidden in the Haystack: Smaller Needles are More Difficult for LLMs to Find
by: Bianchi, Owen, et al.
Published: (2025)

ICLGuard: Controlling In-Context Learning Behavior for Applicability Authorization
by: Si, Wai Man, et al.
Published: (2024)

Regularizing Hidden States Enables Learning Generalizable Reward Model for LLMs
by: Yang, Rui, et al.
Published: (2024)

Synthesizing Post-Training Data for LLMs through Multi-Agent Simulation
by: Tang, Shuo, et al.
Published: (2024)

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

Scalable Defense against In-the-wild Jailbreaking Attacks with Safety Context Retrieval
by: Chen, Taiye, et al.
Published: (2025)

BiLLM: Pushing the Limit of Post-Training Quantization for LLMs
by: Huang, Wei, et al.
Published: (2024)

Jailbreaking Attacks vs. Content Safety Filters: How Far Are We in the LLM Safety Arms Race?
by: Xin, Yuan, et al.
Published: (2025)

Does Safety Training of LLMs Generalize to Semantically Related Natural Prompts?
by: Addepalli, Sravanti, et al.
Published: (2024)

Phi-3 Safety Post-Training: Aligning Language Models with a "Break-Fix" Cycle
by: Haider, Emman, et al.
Published: (2024)

What Is The Political Content in LLMs' Pre- and Post-Training Data?
by: Ceron, Tanise, et al.
Published: (2025)

TeachLM: Post-Training LLMs for Education Using Authentic Learning Data
by: Perczel, Janos, et al.
Published: (2025)

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

GraphDancer: Training LLMs to Explore and Reason over Graphs via Two-Stage Curriculum Post-Training
by: Bai, Yuyang, et al.
Published: (2026)

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

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

A Study on Hidden Layer Distillation for Large Language Model Pre-Training
by: Guigon, Maxime, et al.
Published: (2026)

Can Post-Training Transform LLMs into Causal Reasoners?
by: Chen, Junqi, et al.
Published: (2026)

JailbreakRadar: Comprehensive Assessment of Jailbreak Attacks Against LLMs
by: Chu, Junjie, et al.
Published: (2024)

Examining Reasoning LLMs-as-Judges in Non-Verifiable LLM Post-Training
by: Liu, Yixin, et al.
Published: (2026)

When More is Less: Understanding Chain-of-Thought Length in LLMs
by: Wu, Yuyang, et al.
Published: (2025)

Beyond Reactive Safety: Risk-Aware LLM Alignment via Long-Horizon Simulation
by: Sun, Chenkai, et al.
Published: (2025)

How Post-Training Reshapes LLMs: A Mechanistic View on Knowledge, Truthfulness, Refusal, and Confidence
by: Du, Hongzhe, et al.
Published: (2025)

SignRoundV2: Toward Closing the Performance Gap in Extremely Low-Bit Post-Training Quantization for LLMs
by: Cheng, Wenhua, et al.
Published: (2025)

DynMoLE: Boosting Mixture of LoRA Experts Fine-Tuning with a Hybrid Routing Mechanism
by: Li, Dengchun, et al.
Published: (2025)

EstLLM: Enhancing Estonian Capabilities in Multilingual LLMs via Continued Pretraining and Post-Training
by: Dorkin, Aleksei, et al.
Published: (2026)

JT-Safe: Intrinsically Enhancing the Safety and Trustworthiness of LLMs
by: Feng, Junlan, et al.
Published: (2025)

Finding the Translation Switch: Discovering and Exploiting the Task-Initiation Features in LLMs
by: Wu, Xinwei, et al.
Published: (2026)

LLMs can Find Mathematical Reasoning Mistakes by Pedagogical Chain-of-Thought
by: Jiang, Zhuoxuan, et al.
Published: (2024)

Adaptation of Agentic AI: A Survey of Post-Training, Memory, and Skills
by: Jiang, Pengcheng, et al.
Published: (2025)

KDRL: Post-Training Reasoning LLMs via Unified Knowledge Distillation and Reinforcement Learning
by: Xu, Hongling, et al.
Published: (2025)

How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety by Humanizing LLMs
by: Zeng, Yi, et al.
Published: (2024)

A Primer in Post-Training Reasoning Data: What We Know About How It Works
by: Li, Yaoming, et al.
Published: (2026)

EMBRE: Entity-aware Masking for Biomedical Relation Extraction
by: Li, Mingjie, et al.
Published: (2024)