Cover Image

Safe Transformer: An Explicit Safety Bit For Interpretable And Controllable Alignment

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Feng, Jingyuan, Gambardella, Andrew, Minegishi, Gouki, Kojima, Takeshi, Iwasawa, Yusuke, Matsuo, Yutaka
Format:	Preprint
Published:	2026
Subjects:	Machine Learning Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Towards Empirical Interpretation of Internal Circuits and Properties in Grokked Transformers on Modular Polynomials
by: Furuta, Hiroki, et al.
Published: (2024)

Understanding Emergent Misalignment via Feature Superposition Geometry
by: Minegishi, Gouki, et al.
Published: (2026)

Zipping the Thought: When and How Compressed Reasoning Data Works in LLM Post-Training
by: Matsutani, Kohsei, et al.
Published: (2026)

Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words
by: Minegishi, Gouki, et al.
Published: (2025)

Inconsistent Tokenizations Cause Language Models to be Perplexed by Japanese Grammar
by: Gambardella, Andrew, et al.
Published: (2025)

Topology of Reasoning: Understanding Large Reasoning Models through Reasoning Graph Properties
by: Minegishi, Gouki, et al.
Published: (2025)

Semantic Token Clustering for Efficient Uncertainty Quantification in Large Language Models
by: Cao, Qi, et al.
Published: (2026)

Bridging Lottery Ticket and Grokking: Understanding Grokking from Inner Structure of Networks
by: Minegishi, Gouki, et al.
Published: (2023)

Language Models Do Hard Arithmetic Tasks Easily and Hardly Do Easy Arithmetic Tasks
by: Gambardella, Andrew, et al.
Published: (2024)

RL Squeezes, SFT Expands: A Comparative Study of Reasoning LLMs
by: Matsutani, Kohsei, et al.
Published: (2025)

Beyond Induction Heads: In-Context Meta Learning Induces Multi-Phase Circuit Emergence
by: Minegishi, Gouki, et al.
Published: (2025)

Which Programming Language and What Features at Pre-training Stage Affect Downstream Logical Inference Performance?
by: Uchiyama, Fumiya, et al.
Published: (2024)

C-voting: Confidence-Based Test-Time Voting without Explicit Energy Functions
by: Kubo, Kenji, et al.
Published: (2026)

Residual Koopman Spectral Profiling for Predicting and Preventing Transformer Training Instability
by: Kim, Bum Jun, et al.
Published: (2026)

A Comprehensive Survey on Physical Risk Control in the Era of Foundation Model-enabled Robotics
by: Kojima, Takeshi, et al.
Published: (2025)

Mechanism of Task-oriented Information Removal in In-context Learning
by: Cho, Hakaze, et al.
Published: (2025)

Thinking While Listening: Fast-Slow Recurrence for Long-Horizon Sequential Modeling
by: Takashiro, Shota, et al.
Published: (2026)

ClinDet-Bench: Beyond Abstention, Evaluating Judgment Determinability of LLMs in Clinical Decision-Making
by: Watanabe, Yusuke, et al.
Published: (2026)

$\infty$-MoE: Generalizing Mixture of Experts to Infinite Experts
by: Takashiro, Shota, et al.
Published: (2026)

Self-Harmony: Learning to Harmonize Self-Supervision and Self-Play in Test-Time Reinforcement Learning
by: Wang, Ru, et al.
Published: (2025)

ADOPT: Modified Adam Can Converge with Any $β_2$ with the Optimal Rate
by: Taniguchi, Shohei, et al.
Published: (2024)

GenORM: Generalizable One-shot Rope Manipulation with Parameter-Aware Policy
by: Kuroki, So, et al.
Published: (2023)

Enhancing Unimodal Latent Representations in Multimodal VAEs through Iterative Amortized Inference
by: Oshima, Yuta, et al.
Published: (2024)

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

Omanic: Towards Step-wise Evaluation of Multi-hop Reasoning in Large Language Models
by: Gu, Xiaojie, et al.
Published: (2026)

Interpreting Multi-Attribute Confounding through Numerical Attributes in Large Language Models
by: Takagi, Hirohane, et al.
Published: (2025)

CORA: Conformal Risk-Controlled Agents for Safeguarded Mobile GUI Automation
by: Feng, Yushi, et al.
Published: (2026)

Answer When Needed, Forget When Not: Language Models Pretend to Forget via In-Context Knowledge Unlearning
by: Takashiro, Shota, et al.
Published: (2024)

SafeTuneBed: A Toolkit for Benchmarking LLM Safety Alignment in Fine-Tuning
by: Hossain, Saad, et al.
Published: (2025)

Large Language Models as Theory of Mind Aware Generative Agents with Counterfactual Reflection
by: Yang, Bo, et al.
Published: (2025)

Safe Pruning LoRA: Robust Distance-Guided Pruning for Safety Alignment in Adaptation of LLMs
by: Ao, Shuang, et al.
Published: (2025)

Any-Depth Alignment: Unlocking Innate Safety Alignment of LLMs to Any-Depth
by: Zhang, Jiawei, et al.
Published: (2025)

LittleBit-2: Maximizing the Spectral Energy Gain in Sub-1-Bit LLMs via Latent Geometry Alignment
by: Lee, Banseok, et al.
Published: (2026)

Does "Do Differentiable Simulators Give Better Policy Gradients?'' Give Better Policy Gradients?
by: Onoda, Ku, et al.
Published: (2026)

From Demonstrations to Rewards: Alignment Without Explicit Human Preferences
by: Zeng, Siliang, et al.
Published: (2025)

Curriculum Learning for Safety Alignment
by: Kumar, Sandeep, et al.
Published: (2026)

On Exact Bit-level Reversible Transformers Without Changing Architectures
by: Zhang, Guoqiang, et al.
Published: (2024)

Exposing Limitations of Language Model Agents in Sequential-Task Compositions on the Web
by: Furuta, Hiroki, et al.
Published: (2023)

Interpretable Safety Alignment via SAE-Constructed Low-Rank Subspace Adaptation
by: Wang, Dianyun, et al.
Published: (2025)

SafeCoT: Improving VLM Safety with Minimal Reasoning
by: Ma, Jiachen, et al.
Published: (2025)