Large language models (LLMs) are often deployed to perform constrained tasks, with narrow domains. For example, customer support bots can be built on top of LLMs, relying on their broad language understanding and capabilities to enhance performance. However, these LLMs are adversarially susceptible, potentially generating outputs outside the intended domain. To formalize, assess, and mitigate this risk, we introduce domain certification; a guarantee that accurately characterizes the out-of-domain behavior of language models. We then propose a simple yet effective approach, which we call VALID that provides adversarial bounds as a certificate. Finally, we evaluate our method across a diverse set of datasets, demonstrating that it yields meaningful certificates, which bound the probability of out-of-domain samples tightly with minimum penalty to refusal behavior.
@article{arxiv.2502.19320,
title = {Shh, don't say that! Domain Certification in LLMs},
author = {Cornelius Emde and Alasdair Paren and Preetham Arvind and Maxime Kayser and Tom Rainforth and Thomas Lukasiewicz and Bernard Ghanem and Philip H. S. Torr and Adel Bibi},
journal= {arXiv preprint arXiv:2502.19320},
year = {2025}
}
Comments
10 pages, includes appendix Published in International Conference on Learning Representations (ICLR) 2025