概述
M-25-21 addresses agency use and governance, M-25-22 addresses procurement, and M-26-04 adds AI principles for certain large language models. Agencies should read each memo’s scope and current status rather than treat them as rules for private-sector users.
深入探討
OMB memoranda translate federal policy and statutes into directions for agencies. They can set deadlines, responsibilities and minimum processes, but their audience and scope are defined in each document. An agency should not assume that one AI memo covers every technology, mission or national-security system, and private organizations should not treat federal internal guidance as a general law that directly governs them. OMB Memorandum M-25-21, issued in April 2025, sets out governance, innovation and public-trust requirements for covered agency AI use. It rescinded and replaced the prior M-24-10. It addresses AI developed, used or acquired by or for covered agencies, including agency governance, use-case inventories and additional requirements for high-impact uses. The memo describes high impact in terms of outputs serving as a principal basis for decisions or actions with legal, material, binding or significant effects on rights or safety. The specific implementation rules and exceptions must be read in the memorandum itself. M-25-22, issued the same day, addresses federal AI acquisition and replaced M-24-18. It focuses on how agencies buy AI, including competition, data portability, interoperability and contract terms. M-26-04, dated December 2025, sets Unbiased AI Principles for covered agency use of large language models and complements M-25-21 and M-25-22. It has its own defined agency and model scope and excludes national-security systems as specified in the memo. These documents are related, but their requirements are not identical. The OMB memorandum index is important because guidance can be revised, rescinded or supplemented. A sound summary identifies the memo number, issue date, covered agencies, technology, requirement, exceptions and any later amendment. When a requirement conflicts with another law or policy, agencies must resolve that through authorized legal and policy channels rather than choosing whichever memo is easier. Public versions of inventories and reporting may have their own exceptions. Always consult current official text and agency implementation materials. Legal offices should review edge cases.
戰略影響
風險與安全
災難性和日常的人工智慧危害都取決於誰了解風險以及誰能夠採取行動。
更明確的決策
民眾和專業素養決定強而有力的安全政策在政治上是否可行。
突破炒作
清晰的解釋可以減少炒作、實驗室公關和模糊道德劇場的影響。
The Future of OMB Memos on Federal Agency AI Use
Federal AI guidance may change as OMB issues new memoranda, the administration updates policy and Congress revises statutory requirements. Agencies may refine inventories, high-impact definitions, procurement practices or model-specific principles. National-security uses can follow separate directives. Administrators should maintain a current crosswalk, track rescissions and identify an owner for each requirement rather than relying on a copied checklist. Recheck official guidance at least annually. Policy teams should assign owners for status checks, internal controls and public inventories where required. This work continues.
現實世界的實施
An agency team checks whether an AI use case meets M-25-21’s high-impact definition before applying its additional safeguards.
A procurement officer uses M-25-22 to address vendor documentation, data portability and interoperability in an AI contract.
A governance lead checks M-26-04 scope and its national-security exception before applying its requirements to a model.
A reader checks OMB’s memorandum index for rescissions and updates before quoting an older AI memo.
風險與防護欄
將存在風險視為科幻小說，同時能力複合。
混淆了表面產品安全與高度自治下的對準。
只給非英語和非專業觀眾留下低品質的資源。
實施路線圖
單獨的產品危害、誤用和失控/失調風險。
詢問哪些證據會改變您對時間表和嚴重性的看法。
比起行銷主張，更喜歡主要來源和具體評估。
確定一條行動路徑：職業、政策、資金或技能——而不僅僅是意識。
不斷探索
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the OMB Memos on Federal Agency AI Use quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
常見問題
What is OMB Memos on Federal Agency AI Use?
Office of Management and Budget (OMB) memoranda set requirements and guidance for covered federal agencies that develop, acquire or use AI. M-25-21 addresses agency use and governance, M-25-22 addresses procurement, and M-26-04 adds AI principles for certain large language models. Agencies should read each memo’s scope and current status rather than treat them as rules for private-sector users.
Which memo contains Unbiased AI Principles for certain LLM use?
M-26-04 sets LLM-related principles and complements the 2025 memos.
What should procurement staff consult alongside the memo?
OMB guidance works alongside statutes and agency implementation rules.
How should an agency verify whether an older memo is current?
The index lists current memoranda and later rescissions or updates.
繼續學習
相關指南
為此主題精選的更多指南