Cover Image

Safety Alignment Can Be Not Superficial With Explicit Safety Signals

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Li, Jianwei, Kim, Jung-Eun
Format:	Preprint
Published:	2025
Subjects:	Cryptography and Security Artificial Intelligence Computation and Language Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

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

Trustworthy AI: Safety, Bias, and Privacy -- A Survey
by: Fang, Xingli, et al.
Published: (2025)

Position: Retire the "Positive Backdoor" Label -- Secret Alignment Requires Strict and Systematic Evaluation
by: Li, Jianwei, et al.
Published: (2026)

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

Purifying Generative LLMs from Backdoors without Prior Knowledge or Clean Reference
by: Li, Jianwei, et al.
Published: (2026)

Uncovering, Explaining, and Mitigating the Superficial Safety of Backdoor Defense
by: Min, Rui, et al.
Published: (2024)

Certifying LLM Safety against Adversarial Prompting
by: Kumar, Aounon, et al.
Published: (2023)

On the Role of Attention Heads in Large Language Model Safety
by: Zhou, Zhenhong, et al.
Published: (2024)

An Adversarial Perspective on Machine Unlearning for AI Safety
by: Łucki, Jakub, et al.
Published: (2024)

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

Probing the Robustness of Large Language Models Safety to Latent Perturbations
by: Gu, Tianle, et al.
Published: (2025)

MANATEE: Inference-Time Lightweight Diffusion Based Safety Defense for LLMs
by: Kan, Chun Yan Ryan, et al.
Published: (2026)

Intent Laundering: AI Safety Datasets Are Not What They Seem
by: Golchin, Shahriar, et al.
Published: (2026)

SALAD-Bench: A Hierarchical and Comprehensive Safety Benchmark for Large Language Models
by: Li, Lijun, et al.
Published: (2024)

Bypassing the Safety Training of Open-Source LLMs with Priming Attacks
by: Vega, Jason, et al.
Published: (2023)

Instructional Segment Embedding: Improving LLM Safety with Instruction Hierarchy
by: Wu, Tong, et al.
Published: (2024)

What Matters For Safety Alignment?
by: Li, Xing, et al.
Published: (2026)

A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment
by: Wang, Kun, et al.
Published: (2025)

What is in Your Safe Data? Identifying Benign Data that Breaks Safety
by: He, Luxi, et al.
Published: (2024)

Amplification Effects in Test-Time Reinforcement Learning: Safety and Reasoning Vulnerabilities
by: Khattar, Vanshaj, et al.
Published: (2026)

In Vino Veritas and Vulnerabilities: Examining LLM Safety via Drunk Language Inducement
by: Shetty, Anudeex, et al.
Published: (2026)

Safeguarding Large Language Models in Real-time with Tunable Safety-Performance Trade-offs
by: Fonseca, Joao, et al.
Published: (2025)

Exposing LLM Safety Gaps Through Mathematical Encoding:New Attacks and Systematic Analysis
by: Zhang, Haoyu, et al.
Published: (2026)

Beyond Gradient and Priors in Privacy Attacks: Leveraging Pooler Layer Inputs of Language Models in Federated Learning
by: Li, Jianwei, et al.
Published: (2023)

Challenges in Ensuring AI Safety in DeepSeek-R1 Models: The Shortcomings of Reinforcement Learning Strategies
by: Parmar, Manojkumar, et al.
Published: (2025)

SkillSafetyBench: Evaluating Agent Safety under Skill-Facing Attack Surfaces
by: Jin, Chang, et al.
Published: (2026)

From Threat to Tool: Leveraging Refusal-Aware Injection Attacks for Safety Alignment
by: Chae, Kyubyung, et al.
Published: (2025)

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

Uncovering Logit Suppression Vulnerabilities in LLM Safety Alignment
by: Li, Yuxi, et al.
Published: (2024)

Language Models May Verbatim Complete Text They Were Not Explicitly Trained On
by: Liu, Ken Ziyu, et al.
Published: (2025)

Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonable
by: Huang, Tiansheng, et al.
Published: (2025)

SoSBench: Benchmarking Safety Alignment on Six Scientific Domains
by: Jiang, Fengqing, et al.
Published: (2025)

AgentDoG 1.5: A Lightweight and Scalable Alignment Framework for AI Agent Safety and Security
by: Liu, Dongrui, et al.
Published: (2026)

Secure LLM Fine-Tuning via Safety-Aware Probing
by: Wu, Chengcan, et al.
Published: (2025)

RACC: Representation-Aware Coverage Criteria for LLM Safety Testing
by: Wei, Zeming, et al.
Published: (2026)

ReasAlign: Reasoning Enhanced Safety Alignment against Prompt Injection Attack
by: Li, Hao, et al.
Published: (2026)

Unsafer in Many Turns: Benchmarking and Defending Multi-Turn Safety Risks in Tool-Using Agents
by: Li, Xu, et al.
Published: (2026)

Do We Really Need Curated Malicious Data for Safety Alignment in Multi-modal Large Language Models?
by: Wang, Yanbo, et al.
Published: (2025)

Improving LLM Safety Alignment with Dual-Objective Optimization
by: Zhao, Xuandong, et al.
Published: (2025)

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
by: Hubinger, Evan, et al.
Published: (2024)