Cover Image

Reinforcement Learning Amplifies Emergent Misalignment from Harmless Rewards

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Jørgenvåg, Magnus, Kaczér, David, Ruttert, Lasse, Gülhan, Marvin, Flek, Lucie, Mai, Florian
Format:	Preprint
Published:	2026
Subjects:	Computation and Language
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

In-Training Defenses against Emergent Misalignment in Language Models
by: Kaczér, David, et al.
Published: (2025)

Survey-to-Behavior: Downstream Alignment of Human Values in LLMs via Survey Questions
by: Nie, Shangrui, et al.
Published: (2025)

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

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

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

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)

Encoder Fine-tuning with Stochastic Sampling Outperforms Open-weight GPT in Astronomy Knowledge Extraction
by: Rawat, Shivam, et al.
Published: (2025)

Exploring Robustness of LLMs to Paraphrasing Based on Sociodemographic Factors
by: Arora, Pulkit, et al.
Published: (2025)

Exploring Robustness of Multilingual LLMs on Real-World Noisy Data
by: Aliakbarzadeh, Amirhossein, et al.
Published: (2025)

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

A Critical Reflection and Forward Perspective on Empathy and Natural Language Processing
by: Lahnala, Allison, et al.
Published: (2022)

The Muddy Waters of Modeling Empathy in Language: The Practical Impacts of Theoretical Constructs
by: Lahnala, Allison, et al.
Published: (2025)

Multi-Hop Reasoning for Question Answering with Hyperbolic Representations
by: Welz, Simon, et al.
Published: (2025)

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

Emergent Misalignment is Easy, Narrow Misalignment is Hard
by: Soligo, Anna, et al.
Published: (2026)

Label-Consistent Data Generation for Aspect-Based Sentiment Analysis Using LLM Agents
by: Monfared, Mohammad H. A., et al.
Published: (2026)

Funzac at CoMeDi Shared Task: Modeling Annotator Disagreement from Word-In-Context Perspectives
by: Sarumi, Olufunke O., et al.
Published: (2025)

PERSPECTRA: A Scalable and Configurable Pluralist Benchmark of Perspectives from Arguments
by: Nie, Shangrui, et al.
Published: (2026)

ArithmAttack: Evaluating Robustness of LLMs to Noisy Context in Math Problem Solving
by: Abedin, Zain Ul, et al.
Published: (2025)

More Agents Improve Math Problem Solving but Adversarial Robustness Gap Persists
by: Alavi, Khashayar, et al.
Published: (2025)

Pitfalls of Conversational LLMs on News Debiasing
by: Schlicht, Ipek Baris, et al.
Published: (2024)

Dishonesty in Helpful and Harmless Alignment
by: Huang, Youcheng, et al.
Published: (2024)

Misaligned by Reward: Socially Undesirable Preferences in LLMs
by: Ghazaryan, Gayane, et al.
Published: (2026)

Semantic Containment as a Fundamental Property of Emergent Misalignment
by: Saxena, Rohan
Published: (2026)

Can LLMs Translate Human Instructions into a Reinforcement Learning Agent's Internal Emergent Symbolic Representation?
by: Ma, Ziqi, 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)

Can LLM Agents Identify Spoken Dialects like a Linguist?
by: Bystrich, Tobias, et al.
Published: (2026)

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

LLMs Deceive Unintentionally: Emergent Misalignment in Dishonesty from Misaligned Samples to Biased Human-AI Interactions
by: Hu, Xuhao, et al.
Published: (2025)

Emergent Misalignment via In-Context Learning: Narrow in-context examples can produce broadly misaligned LLMs
by: Afonin, Nikita, et al.
Published: (2025)

Countering Reward Over-optimization in LLM with Demonstration-Guided Reinforcement Learning
by: Rita, Mathieu, et al.
Published: (2024)

Thought Crime: Backdoors and Emergent Misalignment in Reasoning Models
by: Chua, James, 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)

Beyond the Safety Bundle: Auditing the Helpful and Harmless Dataset
by: Chehbouni, Khaoula, et al.
Published: (2024)

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

Unifying the Extremes: Developing a Unified Model for Detecting and Predicting Extremist Traits and Radicalization
by: Lahnala, Allison, et al.
Published: (2025)

How Much of Your Data Can Suck? Thresholds for Domain Performance and Emergent Misalignment in LLMs
by: Ouyang, Jian, et al.
Published: (2025)

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

Do LLMs Provide Consistent Answers to Health-Related Questions across Languages?
by: Schlicht, Ipek Baris, et al.
Published: (2025)

Improving Low-Resource Dialect Classification Using Retrieval-based Voice Conversion
by: Fischbach, Lea, et al.
Published: (2025)