Executive Oversight of Federal AI

America's 2025 AI Action Plan lists 103 federal actions across more than 20 agencies and White House offices, and none has a deadline. The officials who run federal AI hold their roles by executive memo, and Congress's own AI deadlines have slipped for years. Congress should require a public map of who owns each AI function, write key officers into law, set an escalation clock, and demand an annual accounting.

The Problem

Federal agencies are adopting AI fast; at the agencies GAO reviewed, reported uses nearly doubled in a single year.¹ The same systems touch benefits, security, health, and schools, so their risks rarely stay inside one agency.

Responsibility has not kept pace. The 2025 AI Action Plan, signed by three White House officials, lists 103 recommended actions; by our count, none sets a deadline.² Congress created a National AI Initiative Office in 2021, but the statute's core job for it is to "serve as the point of contact" on federal AI activities.³ Three weaknesses follow:

1. Mandates expire. A 2023 executive order created the government's chief AI officers;⁴ it was revoked in January 2025,⁵ and the officers remain because a new OMB memo kept them.⁶ Congress's own AI Center of Excellence at GSA carried a five-year sunset that passed in December 2025.⁷
2. Deadlines slip. Congress gave OMB until September 2021 to issue guidance on agency AI use. In November 2023, GAO found it still missing and warned that key federal AI efforts had "missed deadlines."⁸
3. No one owns the whole. GAO counts ten executive-branch groups with roles in overseeing federal AI, from OSTP and three science-council committees to PCAST. When GAO reviewed them in 2025, OMB and OSTP officials declined to meet with its auditors.⁹

Why legislation: Both parties already agree on part of the fix. The Action Plan itself calls for formalizing the Chief AI Officer Council,² and a bipartisan House bill from Representatives Lieu and Obernolte would write chief AI officers and their council into law.¹⁰ The White House's 2026 legislative framework asks Congress to work through existing regulators instead of creating a new AI rulemaking body,¹¹ which makes it more important to say which agency leads. This is a gap in accountability, not authority: agencies can already act, but no statute says who answers for the whole. When government adopts a technology this consequential, citizens should be able to name the official responsible.

The Solution

A four-step staircase: each step stands alone, and each step up adds accountability. Scope: the government's own AI decisions and cross-agency AI responsibilities, with no new regulator and no new power over private companies; staffing and research are separate proposals.

Step 1 — Publish the map. Within 180 days, OMB and OSTP publish, and update yearly, a map naming the lead and supporting agencies for each major AI function, such as frontier-model security, child safety, procurement, and workforce, along with how disputes are settled and who acts first in a cross-agency incident. A map cannot create authority an agency lacks, but it shows Congress where the gaps are.

Step 2 — Put names on decisions. Write chief AI officers and their council into law, as H.R. 8516 would.¹⁰ For each high-impact government use, one named official signs off on risk assessment, testing, monitoring, and suspension. Make permanent the public inventory of federal AI uses, which current law requires for only five years.¹² Accountability starts with a name.

Step 3 — Start an escalation clock. Give agencies, inspectors general, and protected whistleblowers a secure channel to raise unresolved AI risks. Each escalation gets a named owner, a written decision, and a deadline for follow-through.

Step 4 — Report, and answer for misses. Each year, OMB reports to Congress and the public on plans, deadlines met and missed, material incidents, compliance gaps, and needed authority or funding, with classified annexes where required. Missed statutory deadlines trigger corrective plans and budget review, and inspectors general and GAO keep full access.

Where to start: Step 1 is the floor: it costs little and needs no new power. Step 2 is the heart of the proposal.

Administration and enforcement: OMB issues the framework within 180 days, and agencies implement it over the following year; OSTP keeps research coordination. Enforcement runs through budget review, corrective plans, procurement consequences, and oversight hearings; Congress must define and fund any added binding authority.

Risks and Mitigations

• Another layer of paperwork: Assign one lead per function and retire redundant reports; existing officials can serve as chief AI officers, as OMB already allows.⁶ Coordination will still cost time.
• A regulator by stealth: The statute defines covered decisions and states that the map and the officers gain no power over private companies. Future Congresses will still have to police that line.
• Separation of powers and secrecy: Duties attach to OMB and agency officials rather than the President's personal advisers, and sensitive systems report through classified annexes. Some friction with the White House will remain.

Similar Bills

Fit measures similarity to this proposal's mechanisms: High = direct precedent; Partial = useful component with material differences; Related = adjacent approach.

Federal

State

What this adds: Existing law created a coordinating office and a public inventory, and H.R. 8516 would codify chief AI officers. This proposal adds what none of them provides: a public map of who owns each AI function, an escalation clock for unresolved risks, and an annual accounting of deadlines met and missed.