OMB's AI memos in plain language.
Three Office of Management and Budget memoranda shape how federal agencies use and buy AI today. Here is what each one is for, in a sentence or two.
Wagoner Enterprises · November 2026 · General information, not legal advice. Each claim links to its source.
Why these memos matter
The Office of Management and Budget (OMB) issues memoranda that set policy for federal agencies. Contracting officers, Chief AI Officers, and security teams cite them in solicitations and reviews, usually by number. Knowing which memo covers what makes those requirements easier to read.
M-25-21: how agencies use AI
M-25-21, Accelerating Federal Use of AI through Innovation, Governance, and Public Trust, issued in April 2025, sets how agencies govern the AI they use. It expects each agency to have a Chief AI Officer and sets minimum risk-management practices for AI uses it calls high-impact.
M-25-22: how agencies buy AI
M-25-22, Driving Efficient Acquisition of Artificial Intelligence in Government, issued alongside M-25-21, covers acquisition. It addresses contract terms such as how vendors may use government data and how agencies avoid being locked into one vendor.
M-26-04: buying large language models
M-26-04, Increasing Public Trust in Artificial Intelligence Through Unbiased AI Principles, issued in December 2025, applies to agencies that procure large language models. It implements Executive Order 14319 and sets contract and transparency requirements, which means vendors should expect to document how their models were built and evaluated.
A related change: M-26-05
M-26-05, Adopting a Risk-based Approach to Software and Hardware Security, issued in January 2026, is not an AI memo, but it affects AI software too. It rescinded the single government-wide secure software attestation form set by the earlier memos M-22-18 and M-23-16. Agencies now set their own risk-based requirements, and may still ask for a self-attestation or a software bill of materials.
Where our products fit
The LLM Procurement Compliance Harness organizes the evidence an agency gathers when it buys a large language model. Aegis produces signed evidence of how an AI system's guardrails hold up under test. Both support an agency's own decisions; neither is a substitute for reading the policy.
See what this looks like on your own system.
A pilot runs on your material, inside your boundary, and ends with evidence your assessor can verify.