The Pentagon's AI Oversight Office Is Running on Empty
A GAO audit finds the DoD's responsible-AI division three staffers deep in a ten-seat office — while real-world failures stack up.
By The Defense Desk
The Pentagon office charged with keeping military AI accountable had filled just three of its ten authorized positions as of February 2026 — even as the department pressed forward with a strategy to accelerate AI development across every branch of the armed forces. [1] That staffing gap, documented in a Government Accountability Office report published October 14, is not a footnote. It is the headline.
The GAO found that 29 of 64 efforts outlined in the Defense Department's Responsible AI Strategy and Implementation Pathway remained incomplete as of April 2026. [1] The Responsible AI Working Council — the body meant to coordinate those efforts across DoD components — did not meet at all from late 2024 until August 2025. [1] Officials at the Chief Digital and Artificial Intelligence Office told the watchdog that other priorities, including developing guidance for generative AI, had crowded out the meetings. [1]
Six of the ten DoD components surveyed told the GAO that sharing responsible AI resources and best practices would strengthen their ability to incorporate the department's six Responsible AI tenets into their actual use cases. [1] Without a formal charter defining the council's roles, responsibilities, and meeting cadence, the GAO warned the council risked lapsing again — leaving the department without the collaborative capacity to resolve implementation problems quickly. [1]
The timing matters because the failures are no longer hypothetical. In a September 21 letter to Secretary of Defense Pete Hegseth and Director of National Intelligence Jay Clayton, Senators Mark R. Warner, Jack Reed, and Chris Coons called for Inspector General investigations into two reported incidents: an AI-targeting platform that relied on outdated geospatial data in a kinetic strike in Minab, Iran, and an aborted military interdiction operation driven by potential AI-hallucination in disseminated intelligence by U.S. Special Operations Command Pacific. [2]
[2] They also flagged the Department of Defense's harassment of a leading U.S. AI vendor over the latter's reluctance to allow its products' use in fully-autonomous kinetic operations. [2]
Speed and safety have always competed inside large institutions. But in military AI, the cost of losing that negotiation is not a product recall or a stock dip — it is something considerably harder to walk back.
[1] Without that literacy, service members cannot meaningfully evaluate whether delegating a decision to a machine is appropriate — or when a system has stepped outside its operating boundaries. [1]
The DoD agreed with both GAO recommendations — issuing a charter for the Responsible AI Working Council and addressing the Responsible AI Division's workforce needs. [1] The GAO also urged Congress to order the department to implement earlier recommendations, including defining the responsibilities of organizations involved in AI activities and assigning responsibility and a timeline for identifying and managing the department's AI workforce. [1] CDAO officials, however, told the watchdog they were unlikely to fill vacant positions while the office was being restructured following its realignment to the Office of the Under Secretary of Defense for Research and Engineering. [1]
Agreement and action are different things. Institutional restructuring, competing priorities, and the sheer pace of AI deployment have a way of turning agreed recommendations into perpetual agenda items.
Senator Reed's prescription was plain: "Congress must do its job and provide balanced oversight, standards, and rules of the road." [1] What the GAO audit makes clear is that the road is already being driven — and the guardrails are still on back order.