Pentagon moves to secure military AI

The Pentagon’s AI debate is no longer theoretical. Joint Chiefs Chairman Gen. Dan Caine said on April 23 that autonomous weapons will be a “key and essential part” of future U.S. warfare. (defenseone.com) That remark landed as the Defense Department is pushing AI deeper into command-and-control, drone operations, and routine staff work. A January 9 Pentagon AI strategy ordered the department to become an “AI-first” warfighting force and laid out seven “pace-setting projects.” (media.defense.gov) Some of that expansion is already measurable. A Pentagon official told Breaking Defense that personnel built more than 103,000 semi-autonomous AI agents on GenAI.mil in less than five weeks and logged more than 1.1 million sessions by mid-April. (breakingdefense.com) Those agents are approved only for unclassified work at Impact Level 5, and officials said humans are still supposed to review drafts before anything is submitted. The same report said popular uses include after-action reports, imagery writeups, financial analysis, and planning documents. (breakingdefense.com) The harder question is not whether the military will use AI, but how much judgment stays with people once machines speed up the process. Defense One reported in March that commanders and former officials were warning that heavy AI use can weaken users’ ability to critique outputs and spot false presentations. (defenseone.com) That concern is sharper because one of the Pentagon’s most important AI systems is moving from experiment to institution. Deputy Defense Secretary Steve Feinberg’s March 9 memo put Palantir’s Maven Smart System on a path to become a formal program of record by the end of the fiscal year in September. (defensescoop.com 1) (defensescoop.com 2) Maven is not a chatbot for office work. DefenseScoop described it as an AI-enabled platform that fuses military and intelligence data and compresses the process for finding and striking targets. (defensescoop.com) The Pentagon does have rules on paper. DoD Directive 3000.09, updated in 2023, sets policy for autonomous and semi-autonomous weapon systems and says the department must minimize failures that could lead to unintended engagements. (media.defense.gov) It also has a responsible AI framework. The Pentagon’s Responsible Artificial Intelligence Strategy says the department has to show lawful, ethical, and accountable behavior in designing, testing, procuring, deploying, and using AI so warfighters and leaders can trust the outputs. (media.defense.gov) But watchdogs and outside analysts have warned that governance has not kept pace with adoption. A 2024 Defense Department inspector general report said key implementation documents were overdue and that unclear foundational responsibilities could keep the department from fully achieving AI’s benefits. (media.defense.gov) Pentagon officials have tried to answer that gap with checklists and shared standards. The department’s Responsible AI Toolkit was built as an interactive guide for program managers, and officials said in late 2024 that versions were being mapped for NATO and other partners to help allied systems connect with more confidence. (breakingdefense.com) The result is a military moving on two tracks at once: rapid fielding and slower assurance. The tools are spreading across unclassified networks and operational systems now, while the Pentagon is still trying to prove that humans can keep up with what the software recommends. (breakingdefense.com) (defenseone.com)