OMB AI Memos

The Office of Management and Budget issues the binding policy for how federal agencies use and buy AI. Three memos define the current landscape.

M-25-21 — Minimum risk management for high-impact AI

Issued April 2025 (“Accelerating Federal Use of AI through Innovation, Governance, and Public Trust”). Requires agencies to implement minimum risk-management practices for high-impact AI — systems whose output serves as a principal basis for decisions with legal, material, or binding effects on people’s rights or safety. Minimum practices include: pre-deployment testing, AI impact assessments, adequate human oversight, remedies and appeals for affected individuals, and discontinuing systems that fail the bar.

Deadline: September 22, 2026 — agencies had 18 months. That date has now passed; agencies must report implementation to OMB or discontinue non-compliant systems.

M-25-22 — Federal AI acquisition

Issued April 2025, alongside M-25-21. Sets the rules for how agencies buy AI: responsible procurement practices for AI capabilities, complementing the use-case governance in M-25-21.

M-26-04 — Unbiased AI principles for LLM procurement

Issued December 2025. Implements Executive Order 14319. Requires large language models procured by the government to meet “unbiased AI” principles: truth-seeking (factual accuracy, no ideological distortion) and ideological neutrality (no systematic partisan favor).

Agencies had to update procurement policies by March 11, 2026; new LLM solicitations must include unbiased-AI requirements, with a tiered transparency framework (the top tier requires independent bias evaluations and training-data documentation).

Who they apply to

All federal agencies covered by the memos — which in practice means nearly every agency operating or procuring AI. Contractors should watch closely: requirements flow into solicitations and contracts.