aiqbench analysis
This research introduces CLLPU, a new benchmark addressing the critical challenge of controlling knowledge propagation boundaries during multilingual LLM unlearning. Experiments reveal distinct failure modes where suppression either fails to generalize across languages or leaks beyond intended linguistic scopes, highlighting that general utility can mask damage to neighboring knowledge units. The study establishes precise propagation control as a central requirement for safe and effective multilingual model maintenance.
Beyond Cross-Lingual Transfer: Benchmarking Propagation Boundaries in Multilingual LLM Unlearning
arXiv:2609.05976v1 Announce Type: new Abstract: Large Language Model (LLM) unlearning aims to suppress target knowledge while preserving general capabilities. In multilingual settings, unlearning must additionally propagate within its intended linguistic scope. However, existing evaluations mainly measure cross-lingual transfer and cannot distinguish insufficient from excessive propagation. We introduce CLLPU (Cross-Lingual and Language-Bound Protocol for LLM Unlearning), a multilingual benchmark that formulates this problem through two settings: common-goal forgetting, where target knowledge should be suppressed across all languages, and language-conditioned forgetting, where suppression should remain confined to a designated language. CLLPU combines goal-guided topic pairing, schema-aware relation matching, and dual-anchor multilingual translation to construct 800 matched knowledge-unit pairs and 72,000 QA instances across ten languages. Experiments with six representative methods on Llama-3.1-8B-Instruct reveal opposite failure modes: forgetting remains incomplete when universal suppression is required, yet spreads beyond the intended boundary when language-conditioned confinement is required. We further find that general multilingual utility can conceal damage to neighbor knowledge. These findings establish propagation control as a central challenge for multilingual LLM unlearning. We publicly release CLLPU together with its construction pipeline.
Abstract:Large Language Model (LLM) unlearning aims to suppress target knowledge while preserving general capabilities. In multilingual settings, unlearning must additionally propagate within its intended linguistic scope. However, existing evaluations mainly measure cross-lingual transfer and cannot distinguish insufficient from excessive propagation. We introduce CLLPU (Cross-Lingual and Language-Bound Protocol for LLM Unlearning), a multilingual benchmark that formulates this problem through two settings: common-goal forgetting, where target knowledge should be suppressed across all languages, and language-conditioned forgetting, where suppression should remain confined to a designated language. CLLPU combines goal-guided topic pairing, schema-aware relation matching, and dual-anchor multilingual translation to construct 800 matched knowledge-unit pairs and 72,000 QA instances across ten languages. Experiments with six representative methods on Llama-3.1-8B-Instruct reveal opposite failure modes: forgetting remains incomplete when universal suppression is required, yet spreads beyond the intended boundary when language-conditioned confinement is required. We further find that general multilingual utility can conceal damage to neighbor knowledge. These findings establish propagation control as a central challenge for multilingual LLM unlearning. We publicly release CLLPU together with its construction pipeline.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2609.05976 [cs.CL]
(or arXiv:2609.05976v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.05976 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.
Full text under the publisher's open licence · 271 words
Related coverage
Tiny Aya in the wild: What people are building with open multilingual AI
Cohere · Jul 14, 2026
Benchmarking and Boosting Multilingual Capabilities of LVLMs via OCR-Centric Reinforcement Learning
arXiv · Aug 25, 2026
Who Bridges Safety? Identifying and Targeting Cross-Lingual Shared Safety Pathways
arXiv · Oct 1, 2026
When Benchmarks Mislead: Shortcut Learning, Length Confounds, and the Limits of Cross-Dataset Generalization in Multilingual Fake News and Sarcasm Detection
arXiv · Jul 23, 2026