Editor's note: This article covers the policy baseline as understood from primary sources as of March 19, 2026. Two elements โ the AI Action Plan release and OMB memoranda revision status โ require verification before publication. See review notes in front matter.
Federal agencies deploying AI in 2026 are operating under a policy environment that differs substantially from two years ago. The Biden-era Executive Order 14110 โ "Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence" โ was revoked on January 23, 2025, when President Trump signed a replacement order establishing a new direction for federal AI governance.
Understanding that baseline matters for every federal IT leader who is building a business case, writing an acquisition, or defending an AI deployment to oversight.
What the January 2025 Executive Order Actually Says
The order, "Removing Barriers to American Leadership in Artificial Intelligence," has three operative sections beyond the preamble:
1. Develop an AI Action Plan. Within 180 days (by approximately July 22, 2025), the Assistant to the President for Science and Technology, the Special Advisor for AI and Crypto, and the National Security Advisor were directed to submit an AI Action Plan to the President. The plan's mandate: achieve U.S. AI dominance in ways that "promote human flourishing, economic competitiveness, and national security."
2. Review and roll back Biden-era AI policies. Agencies were directed to identify actions taken under EO 14110 that are inconsistent with the new policy and to suspend, revise, or rescind them. This is a broad mandate โ it includes any agency-level AI policies, directives, or guidance developed in response to the previous order.
3. Revise OMB Memoranda M-24-10 and M-24-18. The OMB Director was directed to revise these memoranda โ which covered AI governance and risk management for federal agencies โ within 60 days of the order (by approximately March 24, 2025).
What This Means in Practice
The revocation of EO 14110 removed the federal AI risk management framework that had been built around it, including the requirement for agencies to designate Chief AI Officers and the associated reporting requirements. What replaced it is still taking shape.
For agencies that were building compliance programs against EO 14110, the transition created real ambiguity. The new order's emphasis is on removing barriers rather than imposing requirements โ which gives agencies more flexibility but also less prescriptive guidance on acceptable AI use, procurement standards, and risk documentation.
This is not inherently a problem. Many agencies were building sensible AI governance frameworks regardless of the specific EO. FedRAMP authorization requirements, FISMA compliance, and existing data governance frameworks don't disappear because an EO was revoked. The practical floor for responsible AI deployment in federal environments hasn't changed: data must be protected, access must be controlled, and systems must be auditable.
The OSTP Role
The Office of Science and Technology Policy continues to operate as the coordinator of federal science and technology policy, including AI. OSTP's director chairs the National Science and Technology Council, which coordinates AI policymaking across the executive branch. For agencies looking for authoritative signals on federal AI direction, OSTP communications remain a primary source.
What Federal IT Leaders Should Be Doing Now
The policy environment in early 2026 rewards agencies that built durable governance frameworks rather than compliance-against-specific-documents. Regardless of which EO is current:
- FedRAMP is still the authorization gate for cloud AI services in federal environments. That hasn't changed and won't.
- FISMA obligations require agencies to assess risk for AI systems that process federal information, just as they would for any other information system.
- Data classification discipline matters more, not less, as AI systems gain the ability to surface and synthesize information across systems.
- Acquisition strategy for AI tools needs to account for model availability constraints in government clouds โ as the Anthropic/GCC situation illustrates, commercial capabilities and government-authorized capabilities are not the same set.
The agencies that will navigate the next two years most effectively are those building AI programs grounded in mission requirements and risk tolerance โ not those optimizing for compliance with whatever document is current.
Primary sources: Executive Order on Removing Barriers to American Leadership in Artificial Intelligence, January 23, 2025; Office of Science and Technology Policy, accessed March 19, 2026; ai.gov, accessed March 19, 2026.