United States - AI Acquisition Policy (M-25-22)

Overview

OMB Memorandum M-25-22, "Driving Efficient Acquisition of Artificial Intelligence in Government," was issued on April 3, 2025 to provide concise, standardized acquisition guidance for AI purchases across Executive Branch agencies. The memorandum rescinds prior procurement guidance on AI and is designed to accelerate federal adoption of AI while protecting privacy, intellectual property, and taxpayer value. It sets out a lifecycle approach to AI procurement (from requirements identification to closeout), requires cross-functional acquisition teams, and instructs agencies to adopt contract terms that reduce vendor lock-in and preserve government rights to data and outputs. The memo also emphasizes a policy preference for American-developed AI and calls for a shared repository of acquisition tools. The official text is published by the White House; see the memorandum at M-25-22 (White House PDF) and the White House fact sheet on the new policies at White House article, Apr 7, 2025.

Definitions

M-25-22 uses procurement- and risk-focused terms that agencies must adopt consistently. Key definitions include: "AI system or service" (software or solution that employs algorithmic or statistical processes to automate, augment, or assist decision-making); "high-impact AI" (uses where AI is a principal basis for agency decisions or where failures would have significant consequences for safety, civil rights, finances, or national security); "cross-functional team" (agency officials from acquisition, technical, privacy, security, legal, programmatic and financial functions); and "performance-based acquisition" (solicitation strategy that specifies outcomes and measurable success metrics). Agencies are instructed to interpret these definitions in light of mission needs and applicable statutory safeguards and to document rationale for high-impact designations in procurement records.

Governance and Institutional Framework

M-25-22 places acquisition governance within existing agency structures while requiring new cross-functional practices. The memorandum directs agency senior leadership (Secretaries, Heads of Agencies) to ensure internal acquisition policies are updated, and it charges Chief Acquisition Officers (CAOs), Chief Information Officers (CIOs), and Chief AI Officers (where appointed under M-25-21) to lead implementation. Agencies must convene cross-functional acquisition teams early—during requirements and market research phases—to identify risks, determine high-impact status, and select appropriate contracting strategies. M-25-22 also directs coordination with the General Services Administration (GSA) and OMB to establish shared procurement tools and to create a web-based repository of AI acquisition tools; agencies are expected to leverage interagency knowledge sharing and the Federal AI community of practice. For official guidance see the memorandum and related GSA implementation pages such as GSA AI Community of Practice.

Key Focus Areas

The memorandum focuses agency attention in several programmatic areas: 1) Risk identification and management—agencies must identify potential risks (privacy, civil rights, safety, performance, supply chain, and model security) early and document mitigation plans; 2) Pre-award testing and validation—where practicable, solicitations should require demonstrations, red-team testing, and pre-award validation for systems that may be high-impact; 3) Contract terms—standardized clauses should address interoperability, data portability, government rights to data and outputs, restrictions on training commercial models with non-public government data, and tailored intellectual property arrangements; 4) Performance monitoring—contracts should include quantitative success metrics, reporting requirements, and continuous monitoring plans; 5) Avoiding vendor lock-in—agencies should require openness, portability, and migration support to preserve competition and long-term value; and 6) Marketplace competition and Buy American preference—agencies are directed to maximize use of American-made AI while remaining consistent with procurement law. These focus areas are designed to protect public interest while enabling timely acquisition of advanced capabilities.

Implementation Framework

Implementation is organized as a lifecycle mapped to procurement phases: (1) Requirements identification (early cross-functional review and initial risk screening); (2) Market research and planning (state-of-the-art assessments, demos, interagency sharing); (3) Solicitation drafting and evaluation (performance requirements, testing plans, IP/data clauses); (4) Pre-award testing and validation (technical evaluation, independent testing where warranted); (5) Contract administration and performance monitoring (metrics, reporting, continuous risk management); and (6) Closeout and lessons learned (performance reviews, data archiving, re-use of assets). Agencies must update internal acquisition regulations, guidance, templates and training to reflect M-25-22. OMB and GSA are tasked to support agencies by providing templates, model contract language, and a centralized repository of acquisition tools and lessons learned.

Monitoring and Evaluation

M-25-22 requires agencies to establish monitoring practices that track AI system performance, compliance with contractual obligations, and emergent risks throughout the system lifecycle. Monitoring includes: routine reporting against defined performance metrics, privacy and civil rights compliance checks, cybersecurity and model-security assessments, and post-deployment auditing. Agencies are expected to incorporate monitoring outputs into acquisition decision-making and to share findings (subject to lawful protections) through interagency channels. OMB oversight and GSA support will facilitate consistency across agencies and enable periodic evaluation of procurement outcomes and market impacts.

Penalties, Liability, and Appeals

As a policy memorandum, M-25-22 does not itself create criminal penalties; compliance is enforced administratively through acquisition and contract mechanisms. Remedies include contractual corrective actions, withholding payments, contract termination for default or convenience, disallowance of costs, administrative debarment or suspension under existing procurement rules where contractor misconduct is found, and referrals to Inspectors General for further action. Agencies retain authority to seek damages or other remedies under the contract. The memorandum also expects agencies to maintain records to support appeals, protests and oversight inquiries under the Competition in Contracting Act and related procurement statutes.

Relationship to Other Instruments

M-25-22 explicitly implements and aligns with broader executive direction and statutory materials. It implements the administration's Executive Order on AI (Executive Order 14179) and complements M-25-21 (AI use policy). M-25-22 rescinds and replaces M-24-18 (the previous AI acquisition memorandum) and should be read alongside applicable procurement law (Federal Acquisition Regulation), privacy and civil rights directives, and relevant sector-specific statutes and guidance (e.g., health, finance). Agencies are required to harmonize their internal policies with M-25-22 and to coordinate with OMB and GSA on changes to acquisition processes.

International Alignment

The memorandum encourages practices that are interoperable with international standards and alignment with allies where appropriate, while emphasizing a policy preference for American-made AI solutions. M-25-22's focus on interoperability, portability, and documented technical requirements supports cross-border interoperability and trading partners' compliance assessments. Agencies are directed to consider export-control obligations, data-transfer safeguards, and international procurement rules when selecting suppliers and drafting contracts. For global policy context and comparative analysis see commentary and OECD and multilateral guidance on public procurement of AI.

Implementation Timeline

Sources and References