Cover Image

When Style Breaks Safety: Defending LLMs Against Superficial Style Alignment

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Xiao, Yuxin, Tonekaboni, Sana, Gerych, Walter, Suriyakumar, Vinith, Ghassemi, Marzyeh
Format:	Preprint
Published:	2025
Subjects:	Machine Learning Artificial Intelligence Computation and Language Computers and Society
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Speak Easy: Eliciting Harmful Jailbreaks from LLMs with Simple Interactions
by: Chan, Yik Siu, et al.
Published: (2025)

Identifying Implicit Social Biases in Vision-Language Models
by: Hamidieh, Kimia, et al.
Published: (2024)

Break the Checkbox: Challenging Closed-Style Evaluations of Cultural Alignment in LLMs
by: Kabir, Mohsinul, et al.
Published: (2025)

Superficial Safety Alignment Hypothesis
by: Li, Jianwei, et al.
Published: (2024)

SFTMix: Elevating Language Model Instruction Tuning with Mixup Recipe
by: Xiao, Yuxin, et al.
Published: (2024)

In the Name of Fairness: Assessing the Bias in Clinical Record De-identification
by: Xiao, Yuxin, et al.
Published: (2023)

KScope: A Framework for Characterizing the Knowledge Status of Language Models
by: Xiao, Yuxin, et al.
Published: (2025)

Defending Against Alignment-Breaking Attacks via Robustly Aligned LLM
by: Cao, Bochuan, et al.
Published: (2023)

An Investigation of Memorization Risk in Healthcare Foundation Models
by: Tonekaboni, Sana, et al.
Published: (2025)

Mind the Style: Impact of Communication Style on Human-Chatbot Interaction
by: Derner, Erik, et al.
Published: (2026)

EigenBench: A Comparative Behavioral Measure of Value Alignment
by: Chang, Jonathn, et al.
Published: (2025)

Learning the Wrong Lessons: Syntactic-Domain Spurious Correlations in Language Models
by: Shaib, Chantal, et al.
Published: (2025)

Evaluation without Generation: Non-Generative Assessment of Harmful Model Specialization with Applications to CSAM
by: Suriyakumar, Vinith M., et al.
Published: (2026)

When Correct Beliefs Collapse: Epistemic Resilience of LLMs under Clinical Pressure
by: Xiao, Boyu, et al.
Published: (2026)

MaskMedPaint: Masked Medical Image Inpainting with Diffusion Models for Mitigation of Spurious Correlations
by: Jin, Qixuan, et al.
Published: (2024)

Hedging and Non-Affirmation: Quantifying LLM Alignment on Questions of Human Rights
by: Javed, Rafiya, et al.
Published: (2025)

Foundational Challenges in Assuring Alignment and Safety of Large Language Models
by: Anwar, Usman, et al.
Published: (2024)

Steering at the Source: Style Modulation Heads for Robust Persona Control
by: Izawa, Yoshihiro, et al.
Published: (2026)

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

UCD: Unlearning in LLMs via Contrastive Decoding
by: Suriyakumar, Vinith M., et al.
Published: (2025)

REQUAL-LM: Reliability and Equity through Aggregation in Large Language Models
by: Ebrahimi, Sana, et al.
Published: (2024)

Revisiting the Superficial Alignment Hypothesis
by: Raghavendra, Mohit, et al.
Published: (2024)

Prompt Attacks Reveal Superficial Knowledge Removal in Unlearning Methods
by: Jang, Yeonwoo, et al.
Published: (2025)

Reaching Beyond the Mode: RL for Distributional Reasoning in Language Models
by: Puri, Isha, et al.
Published: (2026)

PICACO: Pluralistic In-Context Value Alignment of LLMs via Total Correlation Optimization
by: Jiang, Han, et al.
Published: (2025)

Breaking Down Bias: On The Limits of Generalizable Pruning Strategies
by: Ma, Sibo, et al.
Published: (2025)

AXOLOTL: Fairness through Assisted Self-Debiasing of Large Language Model Outputs
by: Ebrahimi, Sana, et al.
Published: (2024)

Surveying Attitudinal Alignment Between Large Language Models Vs. Humans Towards 17 Sustainable Development Goals
by: Wu, Qingyang, et al.
Published: (2024)

StyleRec: A Benchmark Dataset for Prompt Recovery in Writing Style Transformation
by: Liu, Shenyang, et al.
Published: (2025)

Safetywashing: Do AI Safety Benchmarks Actually Measure Safety Progress?
by: Ren, Richard, et al.
Published: (2024)

Generative AI in Medicine
by: Shanmugam, Divya, et al.
Published: (2024)

The MedPerturb Dataset: What Non-Content Perturbations Reveal About Human and Clinical LLM Decision Making
by: Gourabathina, Abinitha, et al.
Published: (2025)

Can LLMs Estimate Student Struggles? Human-AI Difficulty Alignment with Proficiency Simulation for Item Difficulty Prediction
by: Li, Ming, et al.
Published: (2025)

MDAgents: An Adaptive Collaboration of LLMs for Medical Decision-Making
by: Kim, Yubin, et al.
Published: (2024)

Fluent but Foreign: Even Regional LLMs Lack Cultural Alignment
by: Agarwal, Dhruv, et al.
Published: (2025)

Efficient Safety Retrofitting Against Jailbreaking for LLMs
by: Garcia-Gasulla, Dario, et al.
Published: (2025)

Balancing Safety and Helpfulness in Healthcare AI Assistants through Iterative Preference Alignment
by: Nghiem, Huy, et al.
Published: (2025)

"Not in My Backyard": LLMs Uncover Online and Offline Social Biases Against Homelessness
by: Karr Jr., Jonathan A., et al.
Published: (2025)

The Homogenization Problem in LLMs: Towards Meaningful Diversity in AI Safety
by: Rios-Sialer, Ian
Published: (2026)

Are LLMs (Really) Ideological? An IRT-based Analysis and Alignment Tool for Perceived Socio-Economic Bias in LLMs
by: Wachter, Jasmin, et al.
Published: (2025)