OpenAI’s “OpenAI for Government” initiative, announced earlier this week on June 16, combines the company’s recently released ChatGPT Gov offering, custom frontier-AI models for national security, and a $200 million pilot with the U.S. Department of Defense (DoD). It’s a clear signal that Washington sees AI as a strategic imperative, but it also raises urgent questions about bias, hallucinations, data sovereignty, and vendor lock-in.
What OpenAI Brings to Government
OpenAI says the package gives agencies “our best models in locked-down settings.” ChatGPT Gov runs on Microsoft’s Azure Government cloud and meets FedRAMP High and CJIS standards. NASA, Los Alamos and other labs already use it for research. In Minnesota, translators cut hours of work each week. Pennsylvania officials also say they save more than an hour per day on routine tasks.
The crown jewel of the launch is a one-year prototype agreement with the DoD’s Chief Digital and Artificial Intelligence Office (CDAO) to explore AI in administration, military healthcare, and proactive cyber defense.
“This contract, with a $200 million ceiling, will bring OpenAI’s industry-leading expertise to help the Defense Department identify and prototype how frontier AI can transform its administrative operations, from improving how service members and their families get health care, to streamlining how they look at program and acquisition data, to supporting proactive cyber defense” OpenAI said in its announcement.
The heart of the launch is a one-year deal with the DoD’s Chief Digital and Artificial Intelligence Office (CDAO). It will test AI on forms, military medical records and cyber-attack warnings. “This contract will bring OpenAI’s industry-leading expertise to help the Defense Department identify and prototype how frontier AI can transform how it accomplishes its mission in service of the American people,” said Katrina Mulligan, a former Chief of Staff to the Secretary of the Army now at OpenAI.
But the Pentagon’s own AI experts worry. Lt. Gen. (ret.) Jack Shanahan, the former AI chief, warned that AI hallucinations, data-dependencies, and continuous human oversight might be a deal-breaker for intelligence work.
Big Benefits, Clear Dangers
AI can slash hours of data entry, simple analysis and translation. That frees analysts to focus on strategy. It can flag health risks in troop records or spot cyber threats before they hit.
OpenAI chased public-sector deals long before June 2025. In October 2024, Forbes detailed how the company pitched AI tools for battlefield logistics, compliance checks and cyber-defense to the Pentagon. Even then, some defense officials questioned whether the models were ready for sensitive military workflows.
But these solutions also have critical risks. Even locked-down clouds can’t stop the potential for leaked sensitive data. A clever query or a bug might reveal personal or classified info. These AI systems can also sometimes get critical facts wrong. AI sometimes invents answers. In a classified briefing, a made-up detail could derail decisions.
Likewise, the data that AI systems such as OpenAI’s GPT models are trained on may reinforce and exhibit existing data biases. They might misinterpret dialects or favor certain groups over others. Another concern is that of vendor lock-in. Once you build workflows around a proprietary AI, moving to another system costs time and money.
Other Countries Go Their Own Way
Not every government wants U.S. AI in its back office. Singapore is investing S$70 million in SEA-LION, a multilingual, open-source model meant to handle Southeast Asian languages. Brussels is funding OpenEuroLLM and GenAI4EU to produce transparent models in all 24 official EU tongues.
North of the border, more than 5,000 Canadian civil servants are testing CANChat, an in-house chatbot that never leaves national cloud space. Meanwhile, France leans on BLOOM, a public-funded, 176-billion-parameter model that already runs inside several ministries. Taken as a whole, these projects show that AI sovereignty has moved from white papers to day-to-day policy.
Europe is also focused on building its own AI hubs. It has set aside billions for “AI gigafactories,” aiming to host tens of thousands of GPUs. Countries like France back home-grown startups such as Mistral AI. The European AI Act also forces strict checks on any system that affects jobs or legal rights.
India created AIKosha, a data platform for public agencies, and Bhashini, an open-source translator for dozens of local languages to help spread AI tools beyond big cities. Likewise, governments in Japan, New Zealand and Poland are testing their own models, tuned for local laws, languages and services.
Each path shows the same challenge: how to balance the benefits of AI with the requirements for control, openness and trust.
In the coming months, the real test will be whether OpenAI for Government can deliver measurable benefits at the DoD without tripping the ethical landmines that lie beneath.
