Cover Image

From Narrow Unlearning to Emergent Misalignment: Causes, Consequences, and Containment in LLMs

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Mushtaq, Erum, Ramakrishna, Anil, Krishna, Satyapriya, Sahai, Sattvik, Goyal, Prasoon, Chang, Kai-Wei, Zhang, Tao, Gupta, Rahul
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

Not Every Token Needs Forgetting: Selective Unlearning to Limit Change in Utility in Large Language Model Unlearning
by: Wan, Yixin, et al.
Published: (2025)

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

LUME: LLM Unlearning with Multitask Evaluations
by: Ramakrishna, Anil, et al.
Published: (2025)

HARMONY: Hidden Activation Representations and Model Output-Aware Uncertainty Estimation for Vision-Language Models
by: Mushtaq, Erum, et al.
Published: (2025)

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

SemEval-2025 Task 4: Unlearning sensitive content from Large Language Models
by: Ramakrishna, Anil, et al.
Published: (2025)

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

Attribute Controlled Fine-tuning for Large Language Models: A Case Study on Detoxification
by: Meng, Tao, et al.
Published: (2024)

Formation and Evolution of Antimatter Objects
by: Yadav, Sattvik
Published: (2026)

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

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

Assessing Domain-Level Susceptibility to Emergent Misalignment from Narrow Finetuning
by: Mishra, Abhishek, et al.
Published: (2026)

Towards Safety Reasoning in LLMs: AI-agentic Deliberation for Policy-embedded CoT Data Creation
by: Kumarage, Tharindu, et al.
Published: (2025)

ARES: Adaptive Red-Teaming and End-to-End Repair of Policy-Reward System
by: Liang, Jiacheng, et al.
Published: (2026)

Intrinsic Guardrails: How Semantic Geometry of Personality Interacts with Emergent Misalignment in LLMs
by: Aneja, Krishak, et al.
Published: (2026)

Unlearning as multi-task optimization: A normalized gradient difference approach with an adaptive learning rate
by: Bu, Zhiqi, et al.
Published: (2024)

K-Edit: Language Model Editing with Contextual Knowledge Awareness
by: Markowitz, Elan, et al.
Published: (2025)

Amazon Nova AI Challenge -- Trusted AI: Advancing secure, AI-assisted software development
by: Sahai, Sattvik, et al.
Published: (2025)

Adversarial Arena: Crowdsourcing Data Generation through Interactive Competition
by: Goyal, Prasoon, et al.
Published: (2026)

BLUR: A Bi-Level Optimization Approach for LLM Unlearning
by: Reisizadeh, Hadi, et al.
Published: (2025)

Addressing LLM Diversity by Infusing Random Concepts
by: Agrawal, Pulin, et al.
Published: (2026)

Tree-of-Traversals: A Zero-Shot Reasoning Algorithm for Augmenting Black-box Language Models with Knowledge Graphs
by: Markowitz, Elan, et al.
Published: (2024)

Assessing Visual Privacy Risks in Multimodal AI: A Novel Taxonomy-Grounded Evaluation of Vision-Language Models
by: Tsaprazlis, Efthymios, et al.
Published: (2025)

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

Constrained Entropic Unlearning: A Primal-Dual Framework for Large Language Models
by: Entesari, Taha, et al.
Published: (2025)

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

In-Context Explainers: Harnessing LLMs for Explaining Black Box Models
by: Kroeger, Nicholas, et al.
Published: (2023)

Evaluating the Critical Risks of Amazon's Nova Premier under the Frontier Model Safety Framework
by: Krishna, Satyapriya, et al.
Published: (2025)

Understanding the Effects of Iterative Prompting on Truthfulness
by: Krishna, Satyapriya, et al.
Published: (2024)

On the Trade-offs between Adversarial Robustness and Actionable Explanations
by: Krishna, Satyapriya, et al.
Published: (2023)

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

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

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

Towards LLM Unlearning Resilient to Relearning Attacks: A Sharpness-Aware Minimization Perspective and Beyond
by: Fan, Chongyu, et al.
Published: (2025)

SWAN: Semantic Watermarking with Abstract Meaning Representation
by: Ye, Ziping, et al.
Published: (2026)

Rethinking Visual Privacy: A Compositional Privacy Risk Framework for Severity Assessment with VLMs
by: Tsaprazlis, Efthymios, et al.
Published: (2026)

More RLHF, More Trust? On The Impact of Preference Alignment On Trustworthiness
by: Li, Aaron J., et al.
Published: (2024)

Characterizing the Consistency of the Emergent Misalignment Persona
by: Weckauff, Anietta, et al.
Published: (2026)

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

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