Cover Image

CARE: Decoding Time Safety Alignment via Rollback and Introspection Intervention

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Hu, Xiaomeng, Huang, Fei, Yuan, Chenhan, Lin, Junyang, Ho, Tsung-Yi
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

Gradient Cuff: Detecting Jailbreak Attacks on Large Language Models by Exploring Refusal Loss Landscapes
by: Hu, Xiaomeng, et al.
Published: (2024)

Beyond Introspection: Reinforcing Thinking via Externalist Behavioral Feedback
by: Yang, Diji, et al.
Published: (2024)

Nudging: Inference-time Alignment of LLMs via Guided Decoding
by: Fei, Yu, et al.
Published: (2024)

IntroLM: Introspective Language Models via Prefilling-Time Self-Evaluation
by: Kasnavieh, Hossein Hosseini, et al.
Published: (2026)

Zero-Overhead Introspection for Adaptive Test-Time Compute
by: Manvi, Rohin, et al.
Published: (2025)

Latent Phase-Shift Rollback: Inference-Time Error Correction via Residual Stream Monitoring and KV-Cache Steering
by: Gupta, Manan, et al.
Published: (2026)

Test-Time Safety Alignment
by: Saglam, Baturay, et al.
Published: (2026)

Toward Adaptive Reasoning in Large Language Models with Thought Rollback
by: Chen, Sijia, et al.
Published: (2024)

Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications
by: Wei, Boyi, et al.
Published: (2024)

Multilingual Safety Alignment via Self-Distillation
by: Qin, Ruiyang, et al.
Published: (2026)

MESA: Improving MoE Safety Alignment via Decentralized Expertise
by: Sun, Yitong, et al.
Published: (2026)

Introspective Planning: Aligning Robots' Uncertainty with Inherent Task Ambiguity
by: Liang, Kaiqu, et al.
Published: (2024)

Recursive Introspection: Teaching Language Model Agents How to Self-Improve
by: Qu, Yuxiao, et al.
Published: (2024)

Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment
by: Krishna, Kundan, et al.
Published: (2025)

Data Advisor: Dynamic Data Curation for Safety Alignment of Large Language Models
by: Wang, Fei, et al.
Published: (2024)

Interpretable Safety Alignment via SAE-Constructed Low-Rank Subspace Adaptation
by: Wang, Dianyun, et al.
Published: (2025)

RelayLLM: Efficient Reasoning via Collaborative Decoding
by: Huang, Chengsong, et al.
Published: (2026)

Reflect then Learn: Active Prompting for Information Extraction Guided by Introspective Confusion
by: Zhao, Dong, et al.
Published: (2025)

Turning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injection
by: Shairah, Harethah Abu, et al.
Published: (2025)

Towards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neurons
by: Chen, Jianhui, et al.
Published: (2024)

LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety
by: Yang, Junxiao, et al.
Published: (2026)

Lifelong Safety Alignment for Language Models
by: Wang, Haoyu, et al.
Published: (2025)

One-Shot Safety Alignment for Large Language Models via Optimal Dualization
by: Huang, Xinmeng, et al.
Published: (2024)

RIV: Recursive Introspection Mask Diffusion Vision Language Model
by: Li, YuQian, et al.
Published: (2025)

From Introspection to Best Practices: Principled Analysis of Demonstrations in Multimodal In-Context Learning
by: Xu, Nan, et al.
Published: (2024)

Course-Correction: Safety Alignment Using Synthetic Preferences
by: Xu, Rongwu, et al.
Published: (2024)

Evaluating LLM Understanding via Structured Tabular Decision Simulations
by: Li, Sichao, et al.
Published: (2025)

CARE-RFT: Confidence-Anchored Reinforcement Finetuning for Reliable Reasoning in Large Language Models
by: Li, Shuozhe, et al.
Published: (2026)

Unraveling and Mitigating Safety Alignment Degradation of Vision-Language Models
by: Liu, Qin, et al.
Published: (2024)

Direct Alignment of Draft Model for Speculative Decoding with Chat-Fine-Tuned LLMs
by: Goel, Raghavv, et al.
Published: (2024)

SpecDec++: Boosting Speculative Decoding via Adaptive Candidate Lengths
by: Huang, Kaixuan, et al.
Published: (2024)

ComPO: Preference Alignment via Comparison Oracles
by: Chen, Peter, et al.
Published: (2025)

$ϕ$-Decoding: Adaptive Foresight Sampling for Balanced Inference-Time Exploration and Exploitation
by: Xu, Fangzhi, et al.
Published: (2025)

TimeCMA: Towards LLM-Empowered Multivariate Time Series Forecasting via Cross-Modality Alignment
by: Liu, Chenxi, et al.
Published: (2024)

Enhancing Time Series Forecasting via Multi-Level Text Alignment with LLMs
by: Zhao, Taibiao, et al.
Published: (2025)

Model Merging and Safety Alignment: One Bad Model Spoils the Bunch
by: Hammoud, Hasan Abed Al Kader, et al.
Published: (2024)

Emulated Disalignment: Safety Alignment for Large Language Models May Backfire!
by: Zhou, Zhanhui, et al.
Published: (2024)

Safety Alignment Can Be Not Superficial With Explicit Safety Signals
by: Li, Jianwei, et al.
Published: (2025)

Frictive Policy Optimization for LLMs: Epistemic Intervention, Risk-Sensitive Control, and Reflective Alignment
by: Pustejovsky, James, et al.
Published: (2026)

Chain of Execution Supervision Promotes General Reasoning in Large Language Models
by: Chen, Nuo, et al.
Published: (2025)