Salvato in:
Dettagli Bibliografici
Autori principali: FitzGerald, Jack, Bates, Dylan, Lazaridis, Aristotelis, Sharma, Aman, Lu, Vincent, King, Brian, Azami, Yousif, Bailey, Sean, Cao, Jeremy, Damianov, Peter, de Haan, Kevin, Madigan, Joseph, McLaurin, Jeremy, Kerbs, Luke, Tainer, Jonathan, Anderson, Dave, Beck, Jonathan, Cuticello, Jamie, Malkerson, Colton, Saltsman, Tyler
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:https://arxiv.org/abs/2603.10012
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911503910699008
author FitzGerald, Jack
Bates, Dylan
Lazaridis, Aristotelis
Sharma, Aman
Lu, Vincent
King, Brian
Azami, Yousif
Bailey, Sean
Cao, Jeremy
Damianov, Peter
de Haan, Kevin
Madigan, Joseph
McLaurin, Jeremy
Kerbs, Luke
Tainer, Jonathan
Anderson, Dave
Beck, Jonathan
Cuticello, Jamie
Malkerson, Colton
Saltsman, Tyler
author_facet FitzGerald, Jack
Bates, Dylan
Lazaridis, Aristotelis
Sharma, Aman
Lu, Vincent
King, Brian
Azami, Yousif
Bailey, Sean
Cao, Jeremy
Damianov, Peter
de Haan, Kevin
Madigan, Joseph
McLaurin, Jeremy
Kerbs, Luke
Tainer, Jonathan
Anderson, Dave
Beck, Jonathan
Cuticello, Jamie
Malkerson, Colton
Saltsman, Tyler
contents Military Large Language Models (LLMs) must provide accurate information to the warfighter in time-critical and dangerous situations. However, today's LLMs are imbued with safety behaviors that cause the LLM to refuse many legitimate queries in the military domain, particularly those related to violence, terrorism, or military technology. Our gold benchmark for assessing refusal rates, which was developed by veterans of the US Army and special forces, is to our knowledge the first dataset of its kind. We present results for refusal and deflection rates on 31 public models and 3 military models. We observe hard rejection rates as high as 98.2% and soft deflection rates ranging from 0% to 21.3%. We also present results on two additional synthetic datasets and show their correlations with the gold dataset. Finally, we perform abliteration using the Heretic library on a military-tuned gpt-oss-20b model, showing an absolute increase in answer rate of 66.5 points but an average relative decrease of 2% on other military tasks. In our concluding remarks, we argue for deeper specialization, including with mid-training and end-to-end post-training, to achieve zero refusals and maximum military task accuracy for closed military models.
format Preprint
id arxiv_https___arxiv_org_abs_2603_10012
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Measuring and Eliminating Refusals in Military Large Language Models
FitzGerald, Jack
Bates, Dylan
Lazaridis, Aristotelis
Sharma, Aman
Lu, Vincent
King, Brian
Azami, Yousif
Bailey, Sean
Cao, Jeremy
Damianov, Peter
de Haan, Kevin
Madigan, Joseph
McLaurin, Jeremy
Kerbs, Luke
Tainer, Jonathan
Anderson, Dave
Beck, Jonathan
Cuticello, Jamie
Malkerson, Colton
Saltsman, Tyler
Computation and Language
Artificial Intelligence
Military Large Language Models (LLMs) must provide accurate information to the warfighter in time-critical and dangerous situations. However, today's LLMs are imbued with safety behaviors that cause the LLM to refuse many legitimate queries in the military domain, particularly those related to violence, terrorism, or military technology. Our gold benchmark for assessing refusal rates, which was developed by veterans of the US Army and special forces, is to our knowledge the first dataset of its kind. We present results for refusal and deflection rates on 31 public models and 3 military models. We observe hard rejection rates as high as 98.2% and soft deflection rates ranging from 0% to 21.3%. We also present results on two additional synthetic datasets and show their correlations with the gold dataset. Finally, we perform abliteration using the Heretic library on a military-tuned gpt-oss-20b model, showing an absolute increase in answer rate of 66.5 points but an average relative decrease of 2% on other military tasks. In our concluding remarks, we argue for deeper specialization, including with mid-training and end-to-end post-training, to achieve zero refusals and maximum military task accuracy for closed military models.
title Measuring and Eliminating Refusals in Military Large Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2603.10012