Watchdog Finds Pentagon’s Responsible AI Rules Unevenly Followed

The Pentagon is racing to build weapons and tools that use artificial intelligence, but its own guidelines for using the technology responsibly are not always closely followed, according to a report by the Government Accountability Office, the investigative arm of Congress.

The review, released on October 14, surveyed officials from 10 Defense Department components about 10 AI projects, including systems that automatically identify targets on the battlefield, process surveillance data and support tactical decisions. Those projects are among more than 300 AI efforts under way at the department, NPR reported.

The department has six tenets meant to keep military AI responsible, laid out in a 2021 memo. They stress department-wide oversight, testing and verification, real-time monitoring of systems once deployed, and cooperation with allies on shared norms. Alissa Czyz, who directed the review, told NPR the tenets are guidance rather than binding rules.

Most of the teams surveyed had taken at least initial steps to apply the tenets, the GAO found. But the details were uneven. Six of the 10 officials said their teams had identified risks in using their projects in an operational setting. Four said the requirements for their projects were not aligned with the responsible AI tenets, and three said they did not know whether their systems would be continuously monitored after deployment.

The report also found institutional weaknesses. A council meant to coordinate responsible AI work across the department did not meet from late 2024 until August 2025, and officials from four components said they ran into problems during that gap that collaboration could have helped solve. The council has resumed meeting, but it has no charter setting out its members’ roles or how often it meets, which the GAO warned could allow it to lapse again.

The small office that supports the effort is thinly staffed. As of February, only three of the 10 authorized positions in the Responsible AI Division of the Pentagon’s chief digital and AI office had been filled, according to the report, which said the shortage had held back planned work.

The GAO recommended that the defense secretary have the office issue a charter for the coordinating council and address the division’s staffing needs, and it asked Congress to consider directing the department to act on earlier, still-open recommendations about AI coordination. The department agreed with both new recommendations, but a GAO representative told NPR that as of October 22 they had not been carried out.

Defense officials have argued that speed matters more. Defense Secretary Pete Hegseth said in a speech in January that the department’s committee cycles measured in months and years were unacceptable, and that the risks to national security of moving too slowly on military AI outweigh the effects of imperfect alignment. The Pentagon has pledged to become an AI-first fighting force and last month announced a new command focused on autonomous weapons, including drones.

The department did not respond to NPR’s question about when it planned to make the recommended changes.