English

Measuring and Eliminating Refusals in Military Large Language Models

Computation and Language 2026-03-12 v1 Artificial Intelligence

Abstract

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.

Keywords

Cite

@article{arxiv.2603.10012,
  title  = {Measuring and Eliminating Refusals in Military Large Language Models},
  author = {Jack FitzGerald and Dylan Bates and Aristotelis Lazaridis and Aman Sharma and Vincent Lu and Brian King and Yousif Azami and Sean Bailey and Jeremy Cao and Peter Damianov and Kevin de Haan and Joseph Madigan and Jeremy McLaurin and Luke Kerbs and Jonathan Tainer and Dave Anderson and Jonathan Beck and Jamie Cuticello and Colton Malkerson and Tyler Saltsman},
  journal= {arXiv preprint arXiv:2603.10012},
  year   = {2026}
}

Comments

30 pages

R2 v1 2026-07-01T11:13:32.401Z