On September 22, 2026, the clock ran out on one of the most consequential AI directives the federal government has ever issued — and the White House page hosting the memorandum returned a 404 error.
Eighteen months earlier, OMB Memorandum M-25-21, “Accelerating Federal Use of AI through Innovation, Governance, and Public Trust,” gave every federal agency a deadline: implement minimum risk-management practices for high-impact AI, or discontinue those systems.
High-impact AI, as the memo defines it, is any system whose output serves as a principal basis for decisions with legal, material, binding, or significant effects on people’s rights or safety. Think benefits determinations, enforcement targeting, hiring screens. For those systems, agencies had to stand up pre-deployment testing, AI impact assessments, adequate human oversight, and remedies and appeals for affected individuals.
September 22 was the test. Every agency is now supposed to report implementation to OMB — or shut the non-compliant systems down.
Whether agencies actually met the bar is the open question. OMB’s own 2025 inventory counted 3,611 AI use cases across 41 agencies, up 69% from 2024 — the surface area keeps growing while the governance deadline has now passed. Watch for OMB’s compliance readout, inspector-general reviews, and which agencies quietly discontinue systems in the coming months.
The 404 is a fitting symbol: the requirements are real, but the follow-through is where federal AI policy keeps getting tested.