Shh, don't say that! Domain Certification in LLMs
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866912263251689472 |
|---|---|
| author | Emde, Cornelius Paren, Alasdair Arvind, Preetham Kayser, Maxime Rainforth, Tom Lukasiewicz, Thomas Ghanem, Bernard Torr, Philip H. S. Bibi, Adel |
| author_facet | Emde, Cornelius Paren, Alasdair Arvind, Preetham Kayser, Maxime Rainforth, Tom Lukasiewicz, Thomas Ghanem, Bernard Torr, Philip H. S. Bibi, Adel |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_19320 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Shh, don't say that! Domain Certification in LLMs Emde, Cornelius Paren, Alasdair Arvind, Preetham Kayser, Maxime Rainforth, Tom Lukasiewicz, Thomas Ghanem, Bernard Torr, Philip H. S. Bibi, Adel Computation and Language Artificial Intelligence Cryptography and Security Machine Learning 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. |
| title | Shh, don't say that! Domain Certification in LLMs |
| topic | Computation and Language Artificial Intelligence Cryptography and Security Machine Learning |
| url | https://arxiv.org/abs/2502.19320 |