Cover Image

Unforgotten Safety: Preserving Safety Alignment of Large Language Models with Continual Learning

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Alssum, Lama, Itani, Hani, Hammoud, Hasan Abed Al Kader, Torr, Philip, Bibi, Adel, Ghanem, Bernard
Format:	Preprint
Published:	2025
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

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

Beyond the Last Answer: Your Reasoning Trace Uncovers More than You Think
by: Hammoud, Hasan Abed Al Kader, et al.
Published: (2025)

SynthCLIP: Are We Ready for a Fully Synthetic CLIP Training?
by: Hammoud, Hasan Abed Al Kader, et al.
Published: (2024)

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

Forget Less, Retain More: A Lightweight Regularizer for Rehearsal-Based Continual Learning
by: Alssum, Lama, et al.
Published: (2025)

Hala Technical Report: Building Arabic-Centric Instruction & Translation Models at Scale
by: Hammoud, Hasan Abed Al Kader, et al.
Published: (2025)

On Pretraining Data Diversity for Self-Supervised Learning
by: Hammoud, Hasan Abed Al Kader, et al.
Published: (2024)

Reasoning Vectors: Transferring Chain-of-Thought Capabilities via Task Arithmetic
by: Zbeeb, Mohammad, et al.
Published: (2025)

From Categories to Classifiers: Name-Only Continual Learning by Exploring the Web
by: Prabhu, Ameya, et al.
Published: (2023)

Rethinking Safety in LLM Fine-tuning: An Optimization Perspective
by: Kim, Minseon, et al.
Published: (2025)

Train Long, Think Short: Curriculum Learning for Efficient Reasoning
by: Hammoud, Hasan Abed Al Kader, et al.
Published: (2025)

DiffCLIP: Differential Attention Meets CLIP
by: Hammoud, Hasan Abed Al Kader, et al.
Published: (2025)

On the Importance of Pretraining Data Alignment for Atomic Property Prediction
by: Ghunaim, Yasir, et al.
Published: (2025)

An Embarrassingly Simple Defense Against LLM Abliteration Attacks
by: Shairah, Harethah Abu, et al.
Published: (2025)

TAPS: Task Aware Proposal Distributions for Speculative Sampling
by: Zbib, Mohamad, et al.
Published: (2026)

Bi-Factorial Preference Optimization: Balancing Safety-Helpfulness in Language Models
by: Zhang, Wenxuan, et al.
Published: (2024)

AraLingBench A Human-Annotated Benchmark for Evaluating Arabic Linguistic Capabilities of Large Language Models
by: Zbeeb, Mohammad, et al.
Published: (2025)

Towards Interpretable Deep Local Learning with Successive Gradient Reconciliation
by: Yang, Yibo, et al.
Published: (2024)

$β$-CLIP: Text-Conditioned Contrastive Learning for Multi-Granular Vision-Language Alignment
by: Zohra, Fatimah, et al.
Published: (2025)

QuanBench+: A Unified Multi-Framework Benchmark for LLM-Based Quantum Code Generation
by: Slim, Ali, et al.
Published: (2026)

Beyond Linear Probes: Dynamic Safety Monitoring for Language Models
by: Oldfield, James, et al.
Published: (2025)

When Do Prompting and Prefix-Tuning Work? A Theory of Capabilities and Limitations
by: Petrov, Aleksandar, et al.
Published: (2023)

Do as I do (Safely): Mitigating Task-Specific Fine-tuning Risks in Large Language Models
by: Eiras, Francisco, et al.
Published: (2024)

Continual Learning on a Diet: Learning from Sparsely Labeled Streams Under Constrained Computation
by: Zhang, Wenxuan, et al.
Published: (2024)

SAVeS: Steering Safety Judgments in Vision-Language Models via Semantic Cues
by: Hinojosa, Carlos, et al.
Published: (2026)

Can Large Language Model Agents Simulate Human Trust Behavior?
by: Xie, Chengxing, et al.
Published: (2024)

A Guardrail for Safety Preservation: When Safety-Sensitive Subspace Meets Harmful-Resistant Null-Space
by: Zhang, Bingjie, et al.
Published: (2025)

Decoupling Safety into Orthogonal Subspace: Cost-Efficient and Performance-Preserving Alignment for Large Language Models
by: Mou, Yutao, et al.
Published: (2025)

Shh, don't say that! Domain Certification in LLMs
by: Emde, Cornelius, et al.
Published: (2025)

Towards Comprehensive Post Safety Alignment of Large Language Models via Safety Patching
by: Zhao, Weixiang, et al.
Published: (2024)

Universal In-Context Approximation By Prompting Fully Recurrent Models
by: Petrov, Aleksandar, et al.
Published: (2024)

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

Towards Interpretable Sequence Continuation: Analyzing Shared Circuits in Large Language Models
by: Lan, Michael, et al.
Published: (2023)

X-VARS: Introducing Explainability in Football Refereeing with Multi-Modal Large Language Model
by: Held, Jan, et al.
Published: (2024)

Towards Context-Invariant Safety Alignment for Large Language Models
by: Wang, Yixu, et al.
Published: (2026)

SafetyBench: Evaluating the Safety of Large Language Models
by: Zhang, Zhexin, et al.
Published: (2023)

Think in Safety: Unveiling and Mitigating Safety Alignment Collapse in Multimodal Large Reasoning Model
by: Lou, Xinyue, et al.
Published: (2025)

SafeMERGE: Preserving Safety Alignment in Fine-Tuned Large Language Models via Selective Layer-Wise Model Merging
by: Djuhera, Aladin, et al.
Published: (2025)

Safety Alignment of Large Language Models via Contrasting Safe and Harmful Distributions
by: Zhang, Xiaoyun, et al.
Published: (2024)

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