Air Force Shutting Down AI Chatbot NIPRGPT

NIPRGPT, the Air Force’s generative artificial intelligence chatbot, will shut down Dec. 31—making way for the Pentagon’s new GenAI.mil system. 

DefenseScoop first reported the move, and an Air Force spokesperson confirmed it to Air & Space Forces Magazine. 

“NIPRGPT served as an enterprise-grade pathfinder designed to rapidly demonstrate the value of generative AI for the Department of the Air Force,” the spokesperson said. “It was instrumental in collecting real-world usage data for the Department of the Air Force, which also informed the Department of War’s overall GenAI adoption strategy and risk posture.” 

Generative AI produces text, images, software code, and more after the system is “trained” on massive data sets. Within that field, Large Language Models like NIPRGPT are focused specifically on understanding and responding to prompts in a “human-like” way.  

The Air Force rolled out NIPRGPT, developed by the Air Force Research Laboratory, in June 2024. As commercial generative AI systems like ChatGPT and Google’s Gemini exploded in popularity, leaders saw the need to give troops a secure and approved tool, rather than risk them putting sensitive data into unauthorized systems. NIPRGPT was meant specifically for non-classified information. 

Officials say the chatbot proved to be wildly popular. Within three months, AFRL said more than 80,000 Airmen and Guardians had tried it. The Air Force spokesperson said that all told, 700,000 personnel across the entire Pentagon have used it. 

Yet the system also faced controversy. The Army blocked it from its networks in April, citing cybersecurity and data governance concerns. And some AI companies argued the Air Force was wasting time and money by developing its own system rather than leveraging existing commercial ones. 

Now, it’s being decommissioned—but the Pentagon isn’t walking away from generative AI. On Dec. 9, the Department of War unveiled GenAI.mil, a new chatbot meant for the entire department. 

The Air Force spokesperson said NIPRGPT helped pave the way for GenAI.mil. 

“The insights gathered from NIPRGPT were foundational in shaping future requirements, establishing effective guardrails, and defining governance for both Department of War and Department of the Air Force,” the spokesperson said. 

Like NIPRGPT, service members are being encouraged to experiment with GenAI.mil and find ways to make their jobs easier. The system is also not meant for classified data, the Pentagon said, but is certified for controlled unclassified information, making it “secure for operational use.” 

NIPRGPT was based on customized versions of publicly available AI models. The Air Force never publicly disclosed those models, but they reportedly included Google’s Gemini and Anthropic’s Claude. GenAI.mil, on the other hand, launched with just Gemini. Additional models will be incorporated later, the Pentagon promised. 

Defense officials believe generative AI can help service members by automating tasks that are time-consuming or too complex for one person. That could include everything from summarizing massive amounts of intelligence to sorting through acquisition regulations. 

Air Force Maj. Michael Kanaan, then the military deputy CIO of the Pentagon’s Chief Digital and Artificial Intelligence Office, described the potential for generative AI to have a “profound impact” on day-to-day tasks during a Defense Innovation Board meeting last year.

“AI will most impact what is seemingly least compelling from the clickbait headline perspective,” Kanann said. “The most profound AI impacts will inevitably be—whether professionals from all walks of business learn it sooner or later—in the back-office functions, at least in the short term. But it’s an area overlooked for its lack of glamour compared to warfighting applications.”