Guarding Secrets: How Defense Contractors Approach Large Language Models in National Security

James Purtilo discusses the role of AI in terms of national security.
Descriptive image for Guarding Secrets: How Defense Contractors Approach Large Language Models in National Security

Last month, the United States Department of Defense (DoD) announced that it had selected Scale AI to help test and evaluate generative artificial intelligence (AI) for military applications. Generative AI made headlines last year for its ability to create text, images and other data by utilizing algorithmic models in response to a user’s prompt.

This has lead to concerns in what it means for content creators, as well as whether any photos and videos could be trusted as being “real” and not AI-generated. However, generative AI still offers the potential to streamline workflows and review troves of information within seconds – an important capability as the world is increasingly data driven.


This includes harnessing the power of so-called “large language models” (LLMs), which are able to review gargantuan troves of information within seconds and crystallize it into a few key points.

“Any forward-thinking company will be studying the potential for LLMs to improve work flow and productivity in ordinary tasks,” explained James Purtilo, associate professor of computer science at the University of Maryland.

LLMs could certainly have a place in the government’s vast networks of federal contractors, where it could aid in the development of new systems and platforms. Data is power but time is money.


The ability to condense data so quickly could be a benefit for any company, as it could allow teams to be provided with nearly instantaneous pointers – and perhaps even limiting the flow of information on a need-to-know basis.

“Defense contractors would be no different, but they more than most companies must do so with a careful eye on security,” Purtilo told ClearanceJobs. “LLMs might help with, say, understanding an RFP or drafting a proposal, but doing so involves disclosing substantial information so the model has something to work with. Any prompt after that is a new opportunity for the model to volunteer your secrets.”

Click HERE to read the full article

The Department welcomes comments, suggestions and corrections.  Send email to editor [-at-] cs [dot] umd [dot] edu.