M-26-04: The Unbiased AI Principles and What They Mean for LLM Procurement

On December 11, 2025, OMB issued M-26-04, Increasing Public Trust in Artificial Intelligence Through Unbiased AI Principles, fulfilling Section 4 of Executive Order 14319 (July 23, 2025). It complements M-25-22 on the acquisition side and adds a new compliance dimension to federal LLM buying: how the model behaves in its outputs.

The two principles

The E.O. defines two Unbiased AI Principles the memo operationalizes:

Truth-seeking. LLMs “shall be truthful in responding to user prompts seeking factual information or analysis,” prioritizing historical accuracy, scientific inquiry, and objectivity, and acknowledging uncertainty where reliable information is incomplete or contradictory.

Ideological Neutrality. LLMs “shall be neutral, nonpartisan tools that do not manipulate responses in favor of ideological dogmas.” Developers “shall not intentionally encode partisan or ideological judgments” into outputs unless those judgments are prompted by, or readily accessible to, the end user.

Scope and mechanics

The memo applies to any LLM procured by an agency, regardless of how it is deployed, modified, or used afterward. National security systems are excluded (application is encouraged where practicable). Agencies must also weigh defined factors in deciding whether to extend the requirements to LLMs they develop themselves and to non-LLM AI models they procure. Open-source models acquired under free licenses are not automatically covered, but agencies “should establish procedures to perform due diligence” on their alignment with the principles before use.

The operative mechanism is contractual: agencies must build the principles into new solicitations and orders for LLM procurement, with implementation measures detailed in the memo’s appendix, including the documentation agencies should expect from LLM providers about model behavior. For vendors, this lands as compliance representations and documentation obligations attached to how the model was built and how it responds. That is a category of requirement that did not exist in federal contracting two years ago.

Why it matters

Whatever one’s view of the policy substance, M-26-04 confirms a structural trend: the government’s AI requirements are moving from what a system is to how it behaves, and behavior is being made a condition of sale. Model behavior standards in contracts imply testing, documentation, and accountability mechanisms to demonstrate them, and agencies will need evaluation and oversight arrangements to verify them, which is governance infrastructure rather than paperwork. That trajectory, behavior as the governed surface, verified continuously rather than assumed at acceptance, is the same one the Governed Autonomy Framework maps for autonomous agents, where the governed behavior is not just what a model says, but what an agent does. Federal health programs should expect these clauses to become routine in solicitations that touch LLMs at all, and should be building the documentation muscle now.