Some Pentagon teams developing artificial-intelligence tools for battlefield use have not fully followed the department’s responsible-AI guidelines, and several officials could not say whether their systems would be monitored after deployment, a congressional watchdog found. The Government Accountability Office report, issued Oct. 14, examined 10 weapons and other tools that use AI for tasks including identifying targets, processing surveillance and supporting tactical decisions. Six officials involved in the projects said their teams had identified risks to using the systems in operational settings. The projects are part of a wider Pentagon effort involving more than 300 AI initiatives. The Defense Department’s 2021 guidance emphasizes department-wide oversight, testing and real-time monitoring, along with cooperation with allies on shared standards. Four officials told GAO that requirements for their projects were not aligned with the responsible-AI tenets. Three said they did not know whether their systems would be continuously monitored once deployed. The report also noted that some projects were still at an early stage and that most teams had taken at least initial steps toward following the guidance. The oversight office responsible for helping the department implement its AI principles is short-staffed. Although authorized to hire 10 employees in 2023, it had filled only three positions as of February, and the Pentagon was unlikely to fill the vacancies or add staff, GAO said. Review director Alissa Czyz called that a small team for a department-wide responsibility. GAO recommended that the Pentagon clarify the oversight office’s duties and meeting schedule and address its staffing shortage. The department agreed, but had not implemented the recommendations by Oct. 22, according to a GAO representative. The Pentagon did not respond to a request for comment. The findings come as Defense Secretary Pete Hegseth urges the department to accelerate military AI development. In January, the Pentagon announced a goal of becoming an “AI-first” warfighting force. Hegseth has argued that the national-security cost of moving too slowly outweighs the risks of imperfect alignment. The watchdog’s findings underscore the challenge facing the department: moving quickly while maintaining clear responsibility for testing, oversight and the consequences of deploying AI-enabled weapons.