In December 2025, OMB issued Memorandum M-26-04, implementing Executive Order 14319 and establishing “unbiased AI” principles for every large language model the federal government procures or deploys. If you sell LLMs to agencies — or assess them — here’s what changed.
The two principles
Federal LLMs must be: Truth-seeking — outputs should reflect factual accuracy and not suppress or distort information on ideological grounds. Ideologically neutral — models must not systematically favor one political viewpoint, party, or worldview.
What agencies had to do
By March 11, 2026, agencies were required to update procurement policies so new LLM solicitations include unbiased-AI requirements from day one. Solicitations must now request: acceptable use policies, model/system/data cards, end-user resources, and feedback mechanisms. Notably, OMB told agencies not to force vendors to divulge model weights or sensitive technical data.
The transparency tiers
The memo sets a tiered framework, with the top “enhanced transparency” tier allowing agencies to seek pre-training and post-training details, model evaluations, enterprise controls, and disclosure of third-party modifications — plus independent bias evaluations.
Why it matters beyond procurement
This is the first federal attempt to put ideological-neutrality requirements into contract language for AI. DOE has already made compliance material to contract payment and flowed the clause to subcontractors at every tier. Expect the “unbiased AI” clause to show up in LLM solicitations across civilian agencies — and expect vendors’ model cards and evals to become standard due-diligence reading.