The United States Army is employing a prototype generative artificial intelligence tool to identify references to diversity, equity, inclusion, and accessibility (DEIA) for removal from training materials in line with a recent executive order from President Donald Trump.
Officials at the Army’s Training and Doctrine Command (TRADOC)—the major command responsible for training soldiers, developing leaders, and shaping the service’s guidelines, strategies, and concepts—are currently using the AI tool, dubbed CamoGPT, to “review policies, programs, publications, and initiatives for DEIA and report findings,” according to an internal memo reviewed by WIRED.
The memo followed Trump’s signing of a January 27 executive order entitled, “Restoring America’s Fighting Force,” which directed Defense Secretary Pete Hegseth to eliminate all Pentagon policies seen as promoting what that the commander-in-chief declared “un-American, divisive, discriminatory, radical, extremist, and irrational theories” regarding race and gender, a linguistic dragnet that extends as far as past social media posts from official US military accounts.
In an email to WIRED, TRADOC spokesman Army Maj. Chris Robinson confirmed the use of CamoGPT to review DEIA materials.
“[TRADOC] will fully execute and implement all directives outlined in the Executive Orders issued by the President. We ensure that these directives are carried out with the utmost professionalism, efficiency, and in alignment with national security objectives,” Robinson says. “Specific details about internal policies and tactics cannot be discussed. However, the use of all tools in our portfolio, including CamoGPT, to increase productivity at all levels can and will be used.”
Developed last summer to boost productivity and operational readiness across the US Army, CamoGPT currently has around 4,000 users who “interact” with it on a daily basis, Capt. Aidan Doyle, a CamoGPT data engineer, tells WIRED. The tool is used for everything from developing comprehensive training program materials to producing multilingual translations, with TRADOC providing a “proof of concept and demonstration” at last October’s annual Association of the United States Army (AUSA) conference in Washington, DC, according to Robinson.
While Doyle declined to comment on the specifics on how TRADOC officials were likely using the CamoGPT to scan for DEIA-related policies, he described the process of searching through documents as relatively straightforward.
“I would take all the documentation you want to examine, order it all in a collection on CamoGPT, and then ask questions about the documents,” he says. “The way retrieval-augmented generation works is that the more specific your question is to the concepts inside the document, the more detailed information the model will provide back.”
In practical terms, this means that TRADOC officials are likely inputting a large number of documents into CamoGPT and asking the LLM to scan for targeted keywords like “dignity” or “respect” (which, yes, the Army is currently using to screen past digital content) to identify materials for subsequent alteration and bring them in line with Trump’s executive order.
By using CamoGPT, the work of eliminating DEIA-related content will likely result in a rapid change to the US Army’s documentation. “We’re competing with ‘control+F’ in Adobe Acrobat,” Doyle says.
