U.S. Air Force contract to AI2: research on secure large language models for critical applications — a federal contract (USAspending)

Large language models (LLMs — AI models trained on large volumes of text to generate and summarize text, answer questions, and so on) have spread rapidly, but using them in defense and government settings calls for assumptions that differ from ordinary services. The more sensitive the information involved, the more that safety (being "secure") becomes a precondition of use: information entered must not leak, outputs must not mislead decisions with incorrect content, and the system must not be tampered with by third parties. This contract's description positions it as research on such secure LLMs for critical Department of the Air Force applications.

What this contract illustrates is a move by government to partner with an outside AI research organization to secure not AI technology in the abstract, but AI in a form that can be used safely. The awardee, AI2 INCORPORATED, is a private research organization, and the significance lies in how the Air Force as the demand side and an AI research and development provider on the supply side are connected through the framework of a federal contract. What the research and development deliverables will be — the specific functions or outcomes — is not stated in the original description, so nothing definitive can be said here.

Viewed across the broader landscape, this case can be read as one example of the larger trend of government AI procurement. Tracing individual federal contracts recorded in USAspending shows which agency awarded what, to whom, for how much, and for what purpose, making it possible to see how public funds flow toward themes such as secure AI and foundation models. By lining it up against other contracts along axes such as awardee, awarding agency, value, and theme, it can be situated as part of the overall picture of government AI procurement.