GenAI.mil is an official generative artificial intelligence platform launched by the U.S. Department of War last December. It provides secure AI tools to military personnel, civilian employees, and contractors, allowing authorized users to leverage commercial AI capabilities for daily administrative tasks, data analysis, planning, and decision-making while keeping data isolated and secure.
The platform has been adopted across the United States military as the preferred enterprise AI platform. Currently, it has roughly 1.7 million users, including about 500,000 power users, according to a recent DefenseScoop report. Its adoption continues to rise.
“When we came in in the second Trump administration, there were roughly 80,000 people in the department that were using generative AI. We had basically these homebrew, bespoke models that had been developed at AFRL and other places. Now … 1.7 million of our 3 million [personnel] have used GenAI.mil and the frontier models that we have on that,” James Mazol, deputy undersecretary of defense for research and engineering, said Tuesday at the DefenseTalks conference, hosted by DefenseScoop.
At least 100,000 AI agents have been created to aid the Pentagon’s workforce.
“We’re going to proliferate frontier models, not just on our controlled unclassified network, but also SIPR, which is secret, and then JWICS for top secret, and into [Special Access Programs] as well,” he added. “So, we’ve seen this explosion. Of that 1.7 [million], I would say we’ve got about half a million power users, or people that seem to be using generative AI pretty much every day to do their jobs.”
The increased user base highlights that generative AI is moving beyond experimentation and becoming part of the Pentagon’s daily operating environment, suggested Ensar Seker, CISCO at SOCRadar, who told ClearanceJobs, “The important signal is not simply how many people have accessed the platform, but how many are using it regularly to support analysis, software development, logistics, intelligence workflows and administrative functions.”
Such adoption could significantly improve how quickly military and government personnel analyze information and make decisions.
“However, as AI systems become more deeply integrated into sensitive and eventually classified environments, the potential impact of a failure also increases. Access controls, data classification, continuous monitoring, model testing, audit logs, and human accountability must be designed into the platform rather than added later,” Seeker added.
CAC Accessibility
The platform, which is certified at Impact Level 5 (IL5) and authorized to handle Controlled Unclassified Information (CUI) on the NIPRNet (Non-classified Internet Protocol Router Network), can be accessed by verified personnel using a Common Access Card.
GenAI.mil currently hosts enterprise and government-tailored versions of leading AI models, including Google Cloud’s Gemini for Government, OpenAI’s ChatGPT, and xAI’s Grok. The Pentagon has confirmed that user data processed within the platform remains isolated inside the government environment and is not used to train public commercial models.
Cameron Stanley, the Pentagon’s chief digital and AI officer, said at DefenseTalks that the department continues to change how it integrates agents, noting challenges on the horizon.
“AI is getting put into everything. That scares me, mainly because of capacity. We don’t have the capacity. It’s not just us; it’s industry writ large. Capacity for compute is going to be difficult for us to overcome,” Stanley warned. “We need to start being more deliberate about making sure that the models that we have are fit for purpose, that they are the right size and scope and shape for what we’re using them for, so we can reduce the inference side and the inference cost.”
Expanded Use of AI
The Pentagon is on track to build its “AI-first” military force for intelligence analysis, administrative productivity, mission planning, and targeting operations. The technology enables its personnel to comb through massive, disparate datasets, including satellite imagery, social media, data brokers, and drone feeds, and then identify and track objects or people of interest.
“Half a million power users show how quickly AI is becoming part of the Pentagon’s daily work. As these tools move from answering questions to taking action, their boundaries need to be enforced outside the AI itself,” explained Danny Jenkins, CEO of cybersecurity provider ThreatLocker.
However, security concerns remain.
“Access to software, data and tools should be denied by default and granted only when an agent needs it for a specific task,” Jenkins told ClearanceJobs. “Firm controls must also limit what the agent can do once it has that access, even if it receives a flawed instruction or encounters malicious content.”
The use of AI will likely increase within the Pentagon and throughout the federal government.
“The Pentagon is expanding its adoption of and use cases for AI like most large enterprises. They face many of the same challenges only at greater scale,” added Scott Miserendino, chief technology officer at DataBee.
Misrendino told ClearanceJobs that these challenges include understanding and controlling costs as well as providing visibility, transparency and accountability within a shared AI governance model across the services and broader defense industrial base.
“While there is a focus on bringing access to frontier models to their users, this must be balanced with managing the much greater cost and inference capacity needs of these models,” Misrendino continued. “Helping users select the right model for the job will be critical.”
The next stage will likely involve AI agents performing increasingly complex workflows across government systems.
“That creates enormous operational potential, but also introduces risks involving prompt injection, excessive permissions, manipulated data, model supply chains and overreliance on outputs that may appear authoritative but are incorrect,” said Seeker. “The future of military AI will therefore depend not only on deploying the most capable models, but on proving that they remain secure, explainable and controllable under adversarial conditions.”
