Government AI procurement refers to the arrangements through which public agencies acquire and deploy AI systems. In the United States in 2025 the pattern was dominated by concentrated federal procurement arrangements — the General Services Administration's OneGov program and the USAi.gov evaluation platform — alongside agency-specific deployments and a separate, unprocured channel in which individual state legislators adopted consumer AI tools directly.
Federal framework
Federal AI adoption is one of three pillars of America's AI Action Plan, issued July 2025 (see America's AI Action Plan).
The General Services Administration launched its OneGov program in August 2025 (see GSA OneGov Program and USAi Platform (August 2025)). Its components include USAi.gov, a free evaluation platform for federal agencies, and a set of $1-per-agency OneGov deals with frontier developers: Anthropic (Claude for Gov/Enterprise), OpenAI (ChatGPT Enterprise), and Google (Gemini for Government). An AWS OneGov arrangement offered up to $1 billion in federal IT savings, and Box AI signed a workflow-automation deal.
GSA extended the pattern to xAI on September 25, 2025, announcing an agreement making Grok models available to federal agencies at $0.42 per organization. GSA described the terms as its lowest price and longest duration to date and identified xAI as the final frontier model added to its suite of AI offerings. The agreement also provides agencies an upgrade path to Grok enterprise subscriptions aligned with the Federal Risk and Authorization Management Program and Department of Defense Impact Levels, offering expanded features and higher rate limits (Source: gsa.gov). GSA characterizes OneGov as aggregating demand across agencies and standardizing contract terms and pricing (Source: thecgp.org).
Authorization pathway
Nominal-price access is separate from authorization to operate. GSA's acquisition guidance states that all cloud service providers must be FedRAMP-authorized or in the process of obtaining authorization, and that AI cloud services should leverage an existing authority to operate (Source: gsa.gov). FedRAMP has prioritized authorization of AI-based cloud services that provide access to conversational AI engines designed for routine and repeated use by federal workers (Source: fedramp.gov).
Office of Management and Budget cybersecurity official Nick Polk said at a July 28, 2026 summit, which the reporting does not name, that OMB wants to reduce compliance burdens as agencies seek access to frontier models with advanced cyber capabilities, and that it is working "in partnership with the FedRAMP program office to authorize these capabilities" (Source: insideaipolicy.com; body behind a subscriber wall, so this claim rests on the lede).
Procurement rulemaking
In March 2026 GSA proposed a General Services Administration Acquisition Regulation clause governing AI used in the performance of GSA Schedule orders, with a comment deadline of March 20, 2026. Reporting by the Coalition for Government Procurement described the proposal as requiring contractors to use only "American AI Systems" — generally, AI developed and produced in the United States — in performance of Schedule orders, and situated it within a multi-year federal push to steer agencies toward domestically developed AI (Source: thecgp.org).
An analysis by the law firm Gibson Dunn published March 12, 2026 summarized the proposed clause as imposing the following obligations on contractors (Source: gibsondunn.com):
- Use only American AI Systems, excluding components developed or controlled by non-U.S. entities.
- Grant the government an irrevocable license to use the AI system for any lawful purpose.
- Disclose all AI tools used in contract performance, and remain responsible for the compliance of service providers and subcontractors.
- Refrain from using government data to train, fine-tune, or improve models for other customers, and keep government data logically segregated from other customers' data.
- Assign the government ownership of "Custom Developments" — modifications or customizations made to the AI system.
- Ensure the AI system does not refuse to produce data outputs on the basis of the contractor's or service provider's own policies, and supply a tool the analysis describes as neutral rather than favoring particular ideological commitments — a requirement tracking the "unbiased AI principles" of Executive Order 14319 — Preventing Woke AI in the Federal Government discussed below.
- Disclose to the ordering contracting officer any AI systems modified to comply with non-U.S. regulations.
Agency deployments
Federal deployments span law enforcement, defense, health, trade, and regulatory drafting:
- Department of Homeland Security — described as a leading law-enforcement and border AI use case (MIT Technology Review, January 2026).
- Department of Defense — Task Force Lima and the AI Rapid Capabilities Cell (AI RCC), a collaboration between the Chief Digital and Artificial Intelligence Office and the Defense Innovation Unit; see CSET + DOJ + CSIS — US-China AI National-Security Axis (composite source summary).
- Food and Drug Administration — an agency-wide AI rollout plus an AI-Assisted Scientific Review pilot (May 2025 and June 2025).
- Immigration and Customs Enforcement — uses AI from Palantir and OpenAI (The Information, January 29, 2026).
- Customs and Border Protection — an Altana AI contract for Section 232 trade enforcement (October 31, 2025).
- Department of Transportation — reported plans to use Google Gemini to write regulations (ProPublica, January 26, 2026).
- DOGE — developed error-prone AI for review of Department of Veterans Affairs contracts (ProPublica, June 6, 2025).
The organizational arrangements supporting these deployments have themselves shifted. Defense Innovation Unit Director Owen West's decision to narrow DIU's focus from seven portfolio areas to three became public on July 31, 2026 through a memo he shared on social media (Source: insideaipolicy.com; body behind a subscriber wall).
Use-case inventories and compliance
Federal agencies maintain AI use case inventories under Office of Management and Budget reporting requirements. In December 2023, the Government Accountability Office reported roughly 1,200 current or planned agency use cases.
GAO-25-107933 (September 2025) cataloged 94 government-wide AI requirements across 5 laws, 6 executive orders, and 3 guidance documents. Key agency-level obligations as of July 2025 included annual AI use case inventory preparation (ongoing); a Chief AI Officer at every agency; a publicly released AI strategy due September 30, 2025; LLM procurement compliance with the "unbiased AI principles" of EO 14319; and adherence to AI acquisition guidance per OMB M-25-22.
GAO reported an implementation gap between legal requirements and agency compliance: of 35 GAO recommendations issued to 19 agencies in 2023, only 4 had been implemented as of July 2025 (GAO-25-107933: AI Federal Efforts Guided by Requirements and Advisory Groups).
Risks and criticism
On security, Sen. Grassley Letter to CISA re ChatGPT (2026-02-05) documents a CISA Acting Director uploading "for official use only" documents to public ChatGPT.
Public Citizen raised a corporate-capture concern in an August 14, 2025 piece titled "New Federal AI Platform Risks Corporate Capture, Worker Displacement, and Equity Blind Spots."
On accuracy and errors, an NYC AI chatbot was caught telling businesses to break the law and was set to be discontinued by incoming Mayor Mamdani (The Markup, January 30, 2026). xAI's Grok reportedly lost a major government contract after the "MechaHitler" and "white-genocide" output incidents (WIRED, August 14, 2025); the GSA OneGov agreement with xAI followed on September 25, 2025.
Vendor eligibility has also become a subject of litigation. The Department of Defense issued a supply-chain risk designation on March 5, 2026 asserting that Anthropic's products could be used to sabotage U.S. operations, and Anthropic sued four days later; on July 31, 2026 it became public that U.S. District Court Judge Rita Lin had suggested federal agencies' own use of Anthropic's cyber-focused Mythos model could undercut the administration's supply-chain-risk position (Source: insideaipolicy.com; body behind a subscriber wall). The parties, the upstream February 2026 directive and the procedural history are recorded on Anthropic v. United States (Pentagon ban challenge); see also Anthropic Claude Gov + Pentagon Dispute (2025–2026).
State legislators and consumer AI
Reporting in The Economist (April 23, 2026) described a pattern outside the federal-procurement frame: individual state legislators using consumer AI tools — Grok, ChatGPT, Claude — for research, drafting bills, and fact-checking lobbyists in real time. According to a National Conference of State Legislatures survey, 44% of state legislative staff used AI in their work in 2025, up from 20% in 2024 (Source: economist.com).
Documented use cases in that reporting include Kent Roe (R-South Dakota House), who runs first drafts of bills through Grok to refine ideas and check which states have passed similar laws and who prompts Grok to apply a "constitutional stress test"; Roe calls AI "a search engine on steroids" and "an accepted tool, same as a calculator," and cross-runs prompts "ten times or a hundred times over" to calibrate against hallucinations. Nick Hoheisel (R-Kansas House) said research that took hours now takes minutes and that AI hallucinates most on case law. Monique Priestley (D-Vermont House) uses AI to fact-check lobbyists in real time during committee hearings, saying "I've actually caught lobbyists in lies." John McCullough, head of South Dakota's legislative research council, noted that AI-drafted bills are still overseen by lawyers, editors, and researchers, but that those reviewers are themselves overstretched.
The reportage surfaced two structural concerns. The first is the prospect of poorly drafted, hallucination-laced bills overwhelming review staff. The second is critical-thinking displacement; as Hoheisel put it, "I do worry that chatbots take away independent thought. Your constituents aren't electing Claude or ChatGPT. They're electing you."
This channel differs from the GSA-OneGov / USAi federal-procurement track in that it is bottom-up consumer-tool adoption by elected officials, with no procurement layer, no auditing, and no formal vendor relationship. Compliance status under state ethics or FOIA regimes is largely undocumented.
Relationships
- supports: America's AI Action Plan.
- depends-on: Anthropic, OpenAI, Google DeepMind, Amazon.
- instance-of: AI Diffusion
- related: GSA — General Services Administration (AI Deployer), DHS — Department of Homeland Security (AI Deployer), DOD — Department of Defense (AI Deployer), FDA — Food and Drug Administration (AI Deployer), ICE — Immigration and Customs Enforcement (AI Deployer), CBP — Customs and Border Protection (AI Deployer), DOT — Department of Transportation (AI Deployer), DOGE — Department of Government Efficiency (AI Deployer), Defense Innovation Unit (DIU) (agencies whose AI adoption runs through these arrangements), GSA OneGov Program and USAi Platform (August 2025), Sen. Grassley Letter to CISA re ChatGPT (2026-02-05), OMB Memorandum M-24-10, OMB Memorandum M-24-18, Anthropic Claude Gov + Pentagon Dispute (2025–2026) (when procurement breaks down), Anthropic v. United States (Pentagon ban challenge), Executive Order 14319 — Preventing Woke AI in the Federal Government, xAI.