The Department of Defense (DOD) issued new procurement guidance directing acquisition officials to incorporate the Trump administration’s “unbiased AI” principles into contracts for artificial intelligence systems. The guidance includes a provision requiring contractors to respond within 15 business days to written notices of potentially noncompliant outputs.
The Trump administration rebranded DOD as the Department of War.
The guidance, issued last week, provides standardized language for solicitations and contracts, along with documentation requirements intended to help DOD assess whether large language models (LLMs) meet federal standards for truthfulness and ideological neutrality.
The requirements implement President Donald Trump’s Executive Order 14319, Preventing Woke AI in the Federal Government, and Office of Management and Budget (OMB) Memorandum M-26-04, which established governmentwide procurement expectations for LLMs.
“Compliance with the E.O. and OMB Memorandum is mandatory,” John M. Tenaglia, principal director of Defense Pricing, Contracting, and Acquisition Policy, wrote in the guidance.
“All relevant personnel should familiarize themselves with this guidance and apply it to applicable procurements,” he wrote.
However, Tenaglia said DOD components may tailor the model language to individual acquisitions based on their complexity and risk.
That flexibility gives program and contracting officials room to adjust documentation and enforcement provisions while maintaining the underlying policy requirements.
Guidance translates “truth-seeking” and “ideological neutrality”
The policy centers on two principles: “truth-seeking,” which calls for LLMs to prioritize factual accuracy and acknowledge uncertainty, and “ideological neutrality,” which directs models not to manipulate responses in favor of ideological positions.
For DOD acquisition officials, the guidance translates those principles into procurement documentation, solicitation requirements, and contractual provisions that can be used to assess vendor compliance before and after an award. The guidance extends compliance considerations to subcontractors involved in providing covered AI capabilities.
Contractors may be required to provide acceptable use policies, model or system documentation, end-user resources and mechanisms for reporting outputs that appear to violate the principles. Depending on the intended use and risk of an AI system, acquisition officials may also seek information about model training, evaluations, content moderation, safeguards, and other controls affecting model outputs.
The guidance creates a formal process for addressing disputed AI outputs and potential corrective action, including the 15-business-day period for contractors to respond after receiving written notice of an apparent violation.
Compliance provisions are intended to be treated as material contract requirements, giving the department a contractual basis to address violations when a vendor fails to take required corrective action.
DOD included three attachments to help contracting personnel implement the policy. They address selecting appropriate transparency and documentation requirements, standardized solicitation language, and contract provisions that can be incorporated into acquisition documents.
For defense contractors, the guidance makes AI transparency and compliance part of the acquisition process rather than solely a matter of model performance. Vendors seeking DOD business may need to demonstrate how their systems meet the principles, document relevant safeguards, and respond to compliance concerns throughout contract performance.
The guidance follows OMB’s Dec. 11, 2025, memorandum, which requires agencies to incorporate the unbiased AI principles into covered LLM procurements and consider applying similar requirements to other AI models.
OMB also directed agencies to update their procurement policies and procedures by March 11, 2026, and to establish processes for users to report outputs that violate the principles. The DOD guidance does not specify a separate departmentwide implementation deadline.