Imagen de Portada

Safety Anchor: Defending Harmful Fine-tuning via Geometric Bottlenecks

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Lu, Guoxin, Sha, Letian, Wang, Qing, Sun, Peijie, Zhou, Hao, Dai, Hua, Xiao, Fu
Formato:	Preprint
Publicado:	2026
Materias:	Cryptography and Security Artificial Intelligence Computation and Language
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

Mitigating Fine-tuning based Jailbreak Attack with Backdoor Enhanced Safety Alignment
por: Wang, Jiongxiao, et al.
Publicado: (2024)

Safety is Not Only About Refusal: Reasoning-Enhanced Fine-tuning for Interpretable LLM Safety
por: Zhang, Yuyou, et al.
Publicado: (2025)

Defending MoE LLMs against Harmful Fine-Tuning via Safety Routing Alignment
por: Kim, Jaehan, et al.
Publicado: (2025)

Be Your Own Red Teamer: Safety Alignment via Self-Play and Reflective Experience Replay
por: Wang, Hao, et al.
Publicado: (2026)

EnchTable: Unified Safety Alignment Transfer in Fine-tuned Large Language Models
por: Wu, Jialin, et al.
Publicado: (2025)

Fun-tuning: Characterizing the Vulnerability of Proprietary LLMs to Optimization-based Prompt Injection Attacks via the Fine-Tuning Interface
por: Labunets, Andrey, et al.
Publicado: (2025)

Why LLM Safety Guardrails Collapse After Fine-tuning: A Similarity Analysis Between Alignment and Fine-tuning Datasets
por: Hsiung, Lei, et al.
Publicado: (2025)

Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning
por: Huang, Tiansheng, et al.
Publicado: (2024)

Rethinking Bottlenecks in Safety Fine-Tuning of Vision Language Models
por: Ding, Yi, et al.
Publicado: (2025)

Defending Against Weight-Poisoning Backdoor Attacks for Parameter-Efficient Fine-Tuning
por: Zhao, Shuai, et al.
Publicado: (2024)

SPARD: Defending Harmful Fine-Tuning Attack via Safety Projection with Relevance-Diversity Data Selection
por: Chen, Shuhao, et al.
Publicado: (2026)

SafeDecoding: Defending against Jailbreak Attacks via Safety-Aware Decoding
por: Xu, Zhangchen, et al.
Publicado: (2024)

Virus: Harmful Fine-tuning Attack for Large Language Models Bypassing Guardrail Moderation
por: Huang, Tiansheng, et al.
Publicado: (2025)

Jailbreaking Commercial Black-Box LLMs with Explicitly Harmful Prompts
por: Zhang, Chiyu, et al.
Publicado: (2025)

CTRAP: Embedding Collapse Trap to Safeguard Large Language Models from Harmful Fine-Tuning
por: Yi, Biao, et al.
Publicado: (2025)

No Two Devils Alike: Unveiling Distinct Mechanisms of Fine-tuning Attacks
por: Leong, Chak Tou, et al.
Publicado: (2024)

Practical Membership Inference Attacks against Fine-tuned Large Language Models via Self-prompt Calibration
por: Fu, Wenjie, et al.
Publicado: (2023)

HSF: Defending against Jailbreak Attacks with Hidden State Filtering
por: Qian, Cheng, et al.
Publicado: (2024)

HarmLevelBench: Evaluating Harm-Level Compliance and the Impact of Quantization on Model Alignment
por: Belkhiter, Yannis, et al.
Publicado: (2024)

Adversarial Tuning: Defending Against Jailbreak Attacks for LLMs
por: Liu, Fan, et al.
Publicado: (2024)

Deep Research Brings Deeper Harm
por: Chen, Shuo, et al.
Publicado: (2025)

Cross-Task Defense: Instruction-Tuning LLMs for Content Safety
por: Fu, Yu, et al.
Publicado: (2024)

Pharmacist: Safety Alignment Data Curation for Large Language Models against Harmful Fine-tuning
por: Liu, Guozhi, et al.
Publicado: (2025)

SEA: Low-Resource Safety Alignment for Multimodal Large Language Models via Synthetic Embeddings
por: Lu, Weikai, et al.
Publicado: (2025)

AdaSteer: Your Aligned LLM is Inherently an Adaptive Jailbreak Defender
por: Zhao, Weixiang, et al.
Publicado: (2025)

Defending LLM Watermarking Against Spoofing Attacks with Contrastive Representation Learning
por: An, Li, et al.
Publicado: (2025)

GRAID: Synthetic Data Generation with Geometric Constraints and Multi-Agentic Reflection for Harmful Content Detection
por: Rad, Melissa Kazemi, et al.
Publicado: (2025)

Trust No Tool: Evaluating and Defending LLM Agents under Untrusted Tool Feedback
por: Yan, Lecheng, et al.
Publicado: (2026)

Token-level Data Selection for Safe LLM Fine-tuning
por: Li, Yanping, et al.
Publicado: (2026)

SafeAligner: Safety Alignment against Jailbreak Attacks via Response Disparity Guidance
por: Huang, Caishuang, et al.
Publicado: (2024)

Understanding and Mitigating Over-refusal for Large Language Models via Safety Representation
por: Zhang, Junbo, et al.
Publicado: (2025)

SafeReview: Defending LLM-based Review Systems Against Adversarial Hidden Prompts
por: Xin, Yuan, et al.
Publicado: (2026)

HarmChip: Evaluating Hardware Security Centric LLM Safety via Jailbreak Benchmarking
por: Wang, Zeng, et al.
Publicado: (2026)

Segment-Level Coherence for Robust Harmful Intent Probing in LLMs
por: He, Xuanli, et al.
Publicado: (2026)

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

Private Fine-tuning of Large Language Models with Zeroth-order Optimization
por: Tang, Xinyu, et al.
Publicado: (2024)

Iron Sharpens Iron: Defending Against Attacks in Machine-Generated Text Detection with Adversarial Training
por: Li, Yuanfan, et al.
Publicado: (2025)

TWGuard: A Case Study of LLM Safety Guardrails for Localized Linguistic Contexts
por: Chu, Hua-Rong, et al.
Publicado: (2026)

Defend LLMs Through Self-Consciousness
por: Huang, Boshi, et al.
Publicado: (2025)

Privately Learning from Graphs with Applications in Fine-tuning Large Language Models
por: Yin, Haoteng, et al.
Publicado: (2024)