AI Policy Wiki
Dashboard

OMB M-25-22 — Driving Efficient Acquisition of Artificial Intelligence in Government

high confidence · updated 2026-06-06

Trump administration's April 2025 replacement for Biden's M-24-18; provides guidance to federal agencies on responsible AI acquisition, AI lifecycle guidance, and unbiased LLM procurement principles.

OMB M-25-22, "Driving Efficient Acquisition of Artificial Intelligence in Government," is an Office of Management and Budget (OMB) memorandum issued on April 3, 2025 that directs federal agencies on how to buy AI systems and services. It replaced the Biden administration's M-24-18 ("Advancing the Responsible Acquisition of AI in Government," September 2024) and was required by Executive Order 14179 (January 2025). It addresses the acquisition side of federal AI governance, alongside its companion governance memo OMB M-25-21 — Accelerating Federal Use of AI through Innovation, Governance, and Public Trust, issued the same day.

Status and timeline

M-25-22 was issued by OMB on April 3, 2025, the same day as the governance memo OMB M-25-21 — Accelerating Federal Use of AI through Innovation, Governance, and Public Trust. Both memos were required by Executive Order 14179 — Removing Barriers to American Leadership in Artificial Intelligence (January 2025). M-25-22 supersedes OMB Memorandum M-24-18, the September 2024 Biden-era memo on responsible AI acquisition. The memo sets a deadline of September 30, 2025 for agencies to begin adhering to its AI acquisition guidance.

Scope and framing

M-25-22 governs how agencies acquire AI systems, distinct from M-25-21's focus on internal governance. It is framed around efficient acquisition and acceleration of AI adoption while maintaining a set of minimum standards, in contrast to the risk-management and responsible-acquisition framing of the superseded M-24-18. Relative to the Biden memo, M-25-22 adds explicit provisions for generative AI and large language models (LLMs), including ideological-neutrality requirements derived from Executive Order 14319 — Preventing Woke AI in the Federal Government.

Obligations on OMB

The memo tasks OMB with developing additional "playbooks" specific to particular AI types — biometrics, specialized computing infrastructure, and generative AI — highlighting considerations and nuances inherent in those specialized areas, with no stated due date. It also directs OMB to establish and continuously update guidance ensuring that AI acquisition contracts align with statutory requirements and address privacy, data ownership, and security. That guidance obligation, which carries a deadline of December 23, 2024 and a refresh cycle of every two years, derives from the Advancing American AI Act; the December 2024 deadline predates the April 2025 memo and is therefore retroactive.

Obligations on agencies

Agencies are directed to adhere to the AI acquisition guidance by September 30, 2025. Contracts for AI systems or services must address the protection of privacy and government data; the ownership and security of data created or processed by contractors; and the security of training data, algorithms, and components against misuse or alteration.

For procurement of LLMs, M-25-22 incorporates the unbiased AI principles drawn from Executive Order 14319 — Preventing Woke AI in the Federal Government. Contracts to procure LLMs must comply with ideological-neutrality requirements, and vendors must be transparent about their system prompts, specifications, and evaluations. Vendors are not required to disclose model weights.

Relationship to EO 14319

M-25-22 serves as the operational vehicle for implementing the unbiased AI principles of Executive Order 14319 — Preventing Woke AI in the Federal Government in federal procurement. EO 14319 required OMB to issue guidance, with a deadline of November 20, 2025, that accounts for technical limitations in complying with the principles; allows vendor transparency through system-prompt disclosure rather than weight disclosure; avoids over-prescribing so as to permit different approaches to innovation; applies factors for whether to extend the principles to agency-built LLMs versus procured ones; and provides national-security exceptions.

Comparison to M-24-18

DimensionM-24-18 (Biden, Sept 2024)M-25-22 (Trump, Apr 2025)
FramingResponsible acquisition + risk managementEfficient acquisition + acceleration
GenAI-specificGeneral AIExplicit GenAI and LLM provisions
Ideological concernsNone"Unbiased AI principles" (EO 14319 compliance)
PlaybooksNot specifiedPromised for biometrics, GenAI, specialized compute

Relationships