analiza aiqbench
This research proposes a hybrid defense framework that integrates entropy, uncertainty, and geometric features to simultaneously mitigate LLM hallucinations and adversarial vulnerabilities. The approach demonstrates significant gains in both clean-task accuracy and robustness against prompt injections and jailbreak attempts across in-domain and out-of-distribution datasets. For practitioners, the key takeaway is that combining these specific feature types yields superior protection compared to relying on single-feature defenses.
Hybrid Adversarial Defence for Natural Language Understanding Tasks
arXiv:2606.04612v1 Announce Type: new Abstract: Large Language Models (LLMs) are vulnerable both to hallucination and adversarial manipulation. Although these problems are closely related, existing defences typically address them separately. We investigate a hybrid defence framework that combines entropy-based models, designed to reduce hallucinations, with uncertainty-based models and geometric-based models, designed to reduce vulnerability. Under in-domain tests on Natural Language Understanding datasets (FEVER, HotpotQA, CSQA, SIQA) we find our hybrid model improves both clean-task performance (up to 43.34\% increase in accuracy) and adversarial robustness (up to 64.92\% improvement in accuracy and 62.27\% reduction in attack success rate). For out-of-distribution datasets (AeroEngQA, CPIQA) we see similar adversarial robustness from our hybrid model (up to 57.14\% improvement in accuracy). For prompt injection (SafeGuard) and jailbreak detection (AdvBench, DAN) datasets our hybrid model is also very strong (up to 51\% reduction in attack success rate compared to state of the art baseline models). Overall, our results show that combining entropy, uncertainty and geometric features provides a more effective defence strategy than using any single feature alone for both in-domain and out-of-distribution tasks.
Abstract:Large Language Models (LLMs) are vulnerable both to hallucination and adversarial manipulation. Although these problems are closely related, existing defences typically address them separately. We investigate a hybrid defence framework that combines entropy-based models, designed to reduce hallucinations, with uncertainty-based models and geometric-based models, designed to reduce vulnerability. Under in-domain tests on Natural Language Understanding datasets (FEVER, HotpotQA, CSQA, SIQA) we find our hybrid model improves both clean-task performance (up to 43.34\% increase in accuracy) and adversarial robustness (up to 64.92\% improvement in accuracy and 62.27\% reduction in attack success rate). For out-of-distribution datasets (AeroEngQA, CPIQA) we see similar adversarial robustness from our hybrid model (up to 57.14\% improvement in accuracy). For prompt injection (SafeGuard) and jailbreak detection (AdvBench, DAN) datasets our hybrid model is also very strong (up to 51\% reduction in attack success rate compared to state of the art baseline models). Overall, our results show that combining entropy, uncertainty and geometric features provides a more effective defence strategy than using any single feature alone for both in-domain and out-of-distribution tasks.
Computation and Language (cs.CL)
arXiv:2606.04612 [cs.CL]
(or arXiv:2606.04612v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.04612 arXiv-issued DOI via DataCite (pending registration)
Text integral sub licența deschisă a publicației · 192 cuvinte
Subiecte conexe
Research Note: Filtering Subversion-Relevant Information From Pretraining Data Is Feasible
LessWrong · 5 oct. 2026
Assessing Information Ecosystems : How Governments Can Get Ahead of Hybrid Threats
carnegieendowment.org · 5 aug. 2026
Request, Aggregate, Bypass: How Attackers Can Evade LLM Safety Classifiers
CrowdStrike Blog · 6 oct. 2026
Lattice-Based Cryptography Explained
Quantum Zeitgeist · 15 aug. 2026