Washington conditions its AI purchases on viewpoint — the Unbiased AI Principles take effect in 2026

The European Union regulates artificial intelligence by risk: what a system does, how dangerous it is, what safeguards it needs. The United States, in its capacity as the world's largest buyer of information technology, has opened a different front entirely — one that European product law does not contemplate. As of 11 March 2026 every federal agency must write conditions about a model's viewpoint into its purchasing contracts. The instrument is Office of Management and Budget Memorandum M-26-04 (11 December 2025), which implements Executive Order 14319, "Preventing Woke AI in the Federal Government" (23 July 2025).

Two principles, written into the contract

The executive order defines two Unbiased AI Principles that any large language model (LLM) procured by the federal government must satisfy:

• Truth-seeking — LLMs "shall be truthful in responding to user prompts

seeking factual information or analysis," prioritising "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 such as DEI." M-26-04 adds that developers "shall not intentionally encode partisan or ideological judgments into an LLM's outputs unless those judgments are prompted by or otherwise readily accessible to the end user."

These are not aspirations. The executive order directs agencies to write the principles into their contracts as terms, to revise existing contracts where practicable, and — critically — provides that "decommissioning costs shall be charged to the vendor in the event of termination by the agency for the vendor's noncompliance."

What M-26-04 actually requires

The order gave the OMB Director 120 days to issue implementing guidance; the December memorandum, signed by Director Russell T. Vought, is that guidance. It sets out three operative obligations:

• New procurements, immediately. Any solicitation or order for the

procurement of an LLM that an agency issues after the date of the memorandum must include contractual requirements addressing compliance with the Unbiased AI Principles. The same applies to non-LLM AI models where the agency determines the requirements should reach them.

• Procurement policy, by 11 March 2026. No later than that date, agencies

must update their policies and procedures so that LLM contracts carry these clauses — and must build in a process for agency users to report model outputs that violate the principles.

• Compliance as a contract condition. The requirements are to be treated

as material to eligibility and payment under the contract, supporting termination for default where a vendor refuses to take corrective action in identified cases of noncompliance.

To assess compliance, agencies are to obtain documentation sufficient to judge a vendor's risk-management actions at the model, system and application level — while, "where practicable," avoiding demands that compel disclosure of sensitive technical data such as model weights. The memorandum builds on the April 2025 acquisition memorandum M-25-22 and reaches every executive department, military department, independent establishment and wholly-owned government corporation. It does not apply to national security systems (44 U.S.C. 3552(b)(6)), though their operators are encouraged to apply it where practicable.

A procurement mandate, not a market rule

The scope is deliberately narrow in one respect and broad in another. It is narrow because it binds only federal procurement: it imposes no obligation on private-sector AI development or on what a vendor may sell to anyone else. It is broad because the federal government is an anchor customer, and a contractual condition attached to its purchasing reaches up the supply chain to the original model developer — including resellers and intermediary deployers, from whom agencies must still extract enough information to establish compliance. A global laboratory that wants federal business must now be able to attest, through its distribution chain, that its model meets a content standard set in Washington.

Why it matters beyond the US

For organisations tracking transatlantic AI policy, M-26-04 sharpens a contrast already visible in the June 2026 frontier-AI executive order and the state-level patchwork. The EU AI Act and Europe's transparency rules regulate the risk and provenance of a system; they are, by design, viewpoint-neutral and say nothing about whether a model is "ideologically neutral." The US measure runs on a different axis altogether — the content posture of the model — and uses the government's purchasing power rather than a statute to enforce it. The practical upshot for a multinational vendor is that the same LLM may have to satisfy the EU's documentation and systemic-risk duties on one side of the Atlantic and an American truth-seeking-and-neutrality attestation on the other, with termination for default and decommissioning costs as the federal stick. The deadline that made this concrete — 11 March 2026 — has now passed, and the clauses are live in federal solicitations.