Cover Image

In-Training Defenses against Emergent Misalignment in Language Models

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Kaczér, David, Jørgenvåg, Magnus, Vetter, Clemens, Afzal, Esha, Haselhorst, Robin, Flek, Lucie, Mai, Florian
Format:	Preprint
Published:	2025
Subjects:	Machine Learning Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Reinforcement Learning Amplifies Emergent Misalignment from Harmless Rewards
by: Jørgenvåg, Magnus, et al.
Published: (2026)

Superalignment with Dynamic Human Values
by: Mai, Florian, et al.
Published: (2025)

Model Organisms for Emergent Misalignment
by: Turner, Edward, et al.
Published: (2025)

Judging Quality Across Languages: A Multilingual Approach to Pretraining Data Filtering with Language Models
by: Ali, Mehdi, et al.
Published: (2025)

On the Limitations of Language Targeted Pruning: Investigating the Calibration Language Impact in Multilingual LLM Pruning
by: Kurz, Simon, et al.
Published: (2024)

Convergent Linear Representations of Emergent Misalignment
by: Soligo, Anna, et al.
Published: (2025)

Understanding Artificial Theory of Mind: Perturbed Tasks and Reasoning in Large Language Models
by: Nickel, Christian, et al.
Published: (2026)

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

IKnow: Instruction-Knowledge-Aware Continual Pretraining for Effective Domain Adaptation
by: Zhang, Tianyi, et al.
Published: (2025)

Thought Crime: Backdoors and Emergent Misalignment in Reasoning Models
by: Chua, James, et al.
Published: (2025)

Persona-Model Collapse in Emergent Misalignment
by: Costa, Davi Bastos, et al.
Published: (2026)

Shared Parameter Subspaces and Cross-Task Linearity in Emergently Misaligned Behavior
by: Arturi, Daniel Aarao Reis, et al.
Published: (2025)

Decomposing Behavioral Phase Transitions in LLMs: Order Parameters for Emergent Misalignment
by: Arnold, Julian, et al.
Published: (2025)

From Narrow Unlearning to Emergent Misalignment: Causes, Consequences, and Containment in LLMs
by: Mushtaq, Erum, et al.
Published: (2025)

Emergent and Subliminal Misalignment Through the Lens of Data-Mediated Transfer
by: Askin, Baris, et al.
Published: (2026)

Can Stories Help LLMs Reason? Curating Information Space Through Narrative
by: Javadi, Vahid Sadiri, et al.
Published: (2024)

Adaptive Defense against Harmful Fine-Tuning for Large Language Models via Bayesian Data Scheduler
by: Hu, Zixuan, et al.
Published: (2025)

The Devil in the Details: Emergent Misalignment, Format and Coherence in Open-Weights LLMs
by: Dickson, Craig
Published: (2025)

Shifting the Gradient: Understanding How Defensive Training Methods Protect Language Model Integrity
by: Grant, Satchel, et al.
Published: (2026)

USDC: A Dataset of $\underline{U}$ser $\underline{S}$tance and $\underline{D}$ogmatism in Long $\underline{C}$onversations
by: Marreddy, Mounika, et al.
Published: (2024)

Persona Features Control Emergent Misalignment
by: Wang, Miles, et al.
Published: (2025)

Distance-Misaligned Training in Graph Transformers and Adaptive Graph-Aware Control
by: Hou, Qinhan, et al.
Published: (2026)

Overtrained, Not Misaligned
by: Schreiber, Joel, et al.
Published: (2026)

Re-Emergent Misalignment: How Narrow Fine-Tuning Erodes Safety Alignment in LLMs
by: Giordani, Jeremiah
Published: (2025)

Unlocking Emergent Modularity in Large Language Models
by: Qiu, Zihan, et al.
Published: (2023)

Reasoning Primitives in Hybrid and Non-Hybrid LLMs: Do Architectural Differences Yield Advantages in State-Tracking and Recall?
by: Rawat, Shivam, et al.
Published: (2026)

Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs
by: Betley, Jan, et al.
Published: (2025)

Raising Bars, Not Parameters: LilMoo Compact Language Model for Hindi
by: Fatimah, Shiza, et al.
Published: (2026)

Linearization Explains Fine-Tuning in Large Language Models
by: Afzal, Zahra Rahimi, et al.
Published: (2026)

Moloch's Bargain: Emergent Misalignment When LLMs Compete for Audiences
by: El, Batu, et al.
Published: (2025)

From Entropy to Calibrated Uncertainty: Training Language Models to Reason About Uncertainty
by: Jenane, Azza, et al.
Published: (2026)

Chain-of-Defensive-Thought: Structured Reasoning Elicits Robustness in Large Language Models against Reference Corruption
by: Wang, Wenxiao, et al.
Published: (2025)

Adversarial Training for Defense Against Label Poisoning Attacks
by: Bal, Melis Ilayda, et al.
Published: (2025)

Emergent Representations of Program Semantics in Language Models Trained on Programs
by: Jin, Charles, et al.
Published: (2023)

Large Language Models in Cybersecurity: Applications, Vulnerabilities, and Defense Techniques
by: Jaffal, Niveen O., et al.
Published: (2025)

Probing the Robustness of Theory of Mind in Large Language Models
by: Nickel, Christian, et al.
Published: (2024)

BLOCK-EM: Preventing Emergent Misalignment via Latent Blocking
by: Ustaomeroglu, Muhammed, et al.
Published: (2026)

Emergent Low-Rank Training Dynamics in MLPs with Smooth Activations
by: Xu, Alec S., et al.
Published: (2026)

Misalignment from Treating Means as Ends
by: Marklund, Henrik, et al.
Published: (2025)

Emergent Causal-Geometric Dynamics Across Depth in Large Language Models
by: Haim, Shahar, et al.
Published: (2026)