analiza aiqbench
This research introduces Orthogonal Boundary Unlearning (OBU), a mathematical framework designed to surgically remove hazardous capabilities from large language models while preserving their legitimate scientific utility. By constraining parameter updates to an orthogonal null space, the method prevents collateral damage to benign dual-use concepts that typically plagues standard unlearning techniques. The approach demonstrates effective mitigation of adversarial relearning and in-context priming attacks across multiple open-weight foundation models without degrading core model competence.
Orthogonal Boundary Unlearning for Dual-Use Concepts in Large Language Models
arXiv:2610.27490v1 Announce Type: new Abstract: Mitigating catastrophic risks from large language models requires excising hazardous operational capabilities, such as cyberwarfare exploitation and pathogen enhancement, without degrading benign scientific utility. In these dual-use domains, legitimate defensive science and malicious attacks share vocabulary, syntax, and internal model representations, causing standard unlearning methods to induce severe collateral damage (concept bleeding). In this work, we present Orthogonal Boundary Unlearning (OBU), a mathematically grounded framework designed to eliminate collateral damage on benign dual-use capabilities. OBU extracts a domain-matched capability subspace from paired calibration queries via singular value decomposition on empirical activation covariances, and restricts parameter updates strictly to the orthogonal null space. This guarantees that benign representations experience zero parameter perturbation along the spanned calibration subspace. We present two complementary realizations of OBU: Projected LoRA, which applies the orthogonal constraint during iterative gradient-based fine-tuning, and Closed-Form Weight Surgery, which directly edits model weights in a single analytical step without training. Across four modern open-weight foundation models (Gemma-3-4B, Qwen3.5-4B, SmolLM3-3B, and Granite-4.2-3B), OBU achieves strong dual-use concept separation, substantially mitigates in-context priming attacks, and demonstrates durable resistance against adversarial relearning without degrading benign domain competence.
Abstract:Mitigating catastrophic risks from large language models requires excising hazardous operational capabilities, such as cyberwarfare exploitation and pathogen enhancement, without degrading benign scientific utility. In these dual-use domains, legitimate defensive science and malicious attacks share vocabulary, syntax, and internal model representations, causing standard unlearning methods to induce severe collateral damage (concept bleeding). In this work, we present Orthogonal Boundary Unlearning (OBU), a mathematically grounded framework designed to eliminate collateral damage on benign dual-use capabilities. OBU extracts a domain-matched capability subspace from paired calibration queries via singular value decomposition on empirical activation covariances, and restricts parameter updates strictly to the orthogonal null space. This guarantees that benign representations experience zero parameter perturbation along the spanned calibration subspace. We present two complementary realizations of OBU: Projected LoRA, which applies the orthogonal constraint during iterative gradient-based fine-tuning, and Closed-Form Weight Surgery, which directly edits model weights in a single analytical step without training. Across four modern open-weight foundation models (Gemma-3-4B, Qwen3.5-4B, SmolLM3-3B, and Granite-4.2-3B), OBU achieves strong dual-use concept separation, substantially mitigates in-context priming attacks, and demonstrates durable resistance against adversarial relearning without degrading benign domain competence.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2610.27490 [cs.LG]
(or arXiv:2610.27490v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.27490 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 281 cuvinte
Subiecte conexe
UNO: Unlearning via Orthogonalization in Generative models
arXiv · 21 oct. 2026
Fairness-Aware Low-Rank Representation Fine-Tuning
arXiv · 25 aug. 2026
Preserving Utility in Continual LLM Unlearning via Orthogonal Projections
arXiv · 30 oct. 2026
ZeroUnlearn: Few-Shot Knowledge Unlearning in Large Language Models
arXiv · 5 iun. 2026