Newest papers

Safety Mid-Training: Internalizing LLM Safety as a Foundational Capability

Zhengyue Zhao, Liwei Jiang, Yejin Choi, Chaowei Xiao1 · 2026 · arXiv

DOI: 10.48550/arXiv.2610.26809 · open-access PDF

Korean

대형 언어 모델 (LLM)에 대한 안전 조정은 주로 훈련 후에 수행되며, 최근의 작업은 또한 사전 훈련에 안전 데이터를 통합하기 시작했습니다.그러나 안전은 나중에 훈련이 더 이상 안전 데이터를 포함하지 않을 때 악화 될 수 있으며 보안의 이득은 일반적인 능력의 비용으로 종종 발생합니다.이 긴장감은 보안과 일반적인 추론이 별도의 훈련 목표로 취급되어서는 안됩니다. 대신에 보안은 일반적인 작업과 공유 된 추론 구조를 통해 배울 수있는 LLM 능력으로 간주됩니다.

Machine translation (Meta M2M100); the English original follows.

Abstract

Safety alignment for large language models (LLMs) is primarily performed during post-training, while recent work has also begun incorporating safety data into pre-training. Yet safety can degrade when later training no longer includes safety data, and gains in safety often come at the cost of general capability. This tension suggests that safety and general reasoning should not be treated as separate training goals. We instead view safety as an LLM capability that can be learned through a reasoning structure shared with general tasks. Both require the model to understand intent or task goals,

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange