Artificial intelligence (AI) in the United States — overview

Artificial intelligence in the United States refers to the design, development, deployment, and governance of machine-based systems that perform tasks associated with human cognition—recognition, prediction, decision support, and content generation. In practice, “U.S. AI” spans three tightly linked arenas: the commercial ecosystem (companies, capital, talent, infrastructure), the public sector (federal and state policy, procurement, standards), and the research community (universities, national labs, and consortia).

As of September 2025, the United States does not have a single comprehensive federal AI statute. Instead, governance looks like a mosaic: executive actions, Office of Management and Budget (OMB) guidance for federal agencies, standards work led by NIST, sectoral and procurement rules, and a growing body of state laws.

What this overview includes—and what it doesn’t

• Includes: Core AI technologies (machine learning, deep learning, foundation models), enabling infrastructure (compute, data centers, and advanced semiconductors), workforce and education, safety and risk management practices, and the evolving federal–state policy landscape.
• Does not include: A single “AI law” binding nationwide; science‑fiction notions of artificial general intelligence as a mainstream policy object; or vendor marketing claims positioned as governance.

Think of the system as an ecosystem more than a codebook: policy signals shape standards and procurement, which influence how companies build and audit systems, which in turn inform research priorities and public debate.

Federal policy: the high-level arc

Two recent moments anchor the current conversation:

• In 2023, the federal government issued a wide‑ranging executive order outlining principles for safety, security, civil rights, and standards coordination, published as the official text in the Federal Register: Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (2023, Federal Register).
• In 2025, policy direction pivoted toward accelerating innovation and infrastructure via a whole‑of‑government plan identifying over 90 actions across pillars such as AI infrastructure and international leadership; see the White House’s PDF: America’s AI Action Plan (July 2025, White House).

Between and beyond these anchor documents, practical governance flows through agency implementation. For example, OMB has directed federal agencies to inventory AI use cases, designate senior AI leaders, and apply risk management practices to safety‑ or rights‑impacting systems—expect these requirements to appear in solicitations and vendor due diligence. The Department of Commerce, through the National Institute of Standards and Technology (NIST), continues to steer voluntary standards and evaluation guidance used across sectors.

Standards and safety: NIST’s central role

The National Institute of Standards and Technology provides widely referenced, voluntary guidance for trustworthy AI. Its core document, the AI Risk Management Framework (AI RMF), helps organizations manage AI risk through four functions—Govern, Map, Measure, and Manage—designed to be adapted to different sectors and system types. For details and the latest resources, see the NIST AI Risk Management Framework main page.

Practically, many U.S. organizations align their AI program to the AI RMF by:

• Defining roles and escalation paths (Govern)
• Mapping use cases, stakeholders, and context of use (Map)
• Establishing evaluation metrics, red‑team protocols, and monitoring (Measure)
• Implementing mitigations, documenting decisions, and setting feedback loops (Manage)

NIST also supports evaluation initiatives—such as domain challenges and guidance on testing generative systems—that reinforce a lifecycle approach: pre‑deployment testing, operational monitoring, and incident response.

State-level activity: a fast-moving patchwork

State legislatures continue to pursue targeted AI rules, including transparency measures for generative systems, safeguards against algorithmic discrimination, and requirements for government AI use. Because bills and implementation timelines vary, operators should monitor the latest session developments and enacted laws using authoritative trackers like the NCSL: Artificial Intelligence 2025 Legislation.

A common operational implication: multi‑state businesses may need jurisdictional matrices capturing disclosure duties, impact assessment triggers, and documentation expectations for high‑risk use cases (for example, employment screening).

Research and infrastructure: broadening access

Beyond policy and standards, the U.S. invests in research capacity. The National AI Research Resource (NAIRR) is a pilot effort to expand access to compute, datasets, software tools, and training, especially for researchers and educators who lack large‑scale infrastructure. For current status and participation pathways, consult the National Artificial Intelligence Research Resource pilot overview (NSF).

This focus on access complements private investment in data centers and advanced semiconductors to support training and inference for increasingly capable models.

How this plays out for organizations

Whether you are a public agency, a federal supplier, or a commercial operator, the practical touchpoints tend to converge:

• Governance and roles: Name accountable owners (e.g., Chief AI Officer or equivalent), clarify decision rights, and set intake/approval processes for AI use cases.
• Use‑case mapping: Identify intended use, affected stakeholders, possible harms, and applicable laws or sectoral rules.
• Evaluation and testing: Define metrics aligned to the system’s purpose, conduct pre‑deployment and ongoing tests (including adversarial or red‑team exercises), and document results.
• Data governance: Establish provenance, quality checks, retention limits, and access controls, especially for training data and model outputs.
• Human oversight: Specify when and how people can override or review system outputs, and train users for appropriate use.
• Monitoring and incidents: Instrument production systems to detect degradation or misuse; create an incident response playbook and reporting channels.
• Procurement alignment: Expect federal and some state buyers to ask for AI use‑case inventories, impact assessments, and evidence of alignment with recognized frameworks (for example, NIST AI RMF).

Common misconceptions, clarified

• “There is one national AI law.” Not as of September 2025. The U.S. approach combines executive actions, agency guidance, standards, procurement, and state laws.
• “AI policy is only about banning or approving models.” Most activity focuses on risk‑based management: evaluation, transparency, human oversight, data governance, and context‑specific safeguards.
• “Policy only applies to frontier models.” Many requirements and best practices apply to any system that meaningfully affects health, safety, civil rights, or critical functions, regardless of model size.

Quick FAQ

• What should a U.S. company follow today? Start with a documented AI program mapped to the NIST AI RMF. Track sectoral obligations (e.g., financial services, health), and watch for state‑level duties like disclosures or impact assessments.
• Does federal guidance apply to vendors? Federal agency requirements often appear in procurement language. If you sell to government, expect requests for your AI inventory, risk assessments, and incident processes.
• How are deepfakes handled? States and federal entities are experimenting with transparency requirements, content provenance signals, and labeling in specific contexts. Prepare governance for synthetic content creation and detection.
• What about export controls and chips? The Department of Commerce has refined controls on advanced computing and semiconductor items since 2023. If you build or deploy models with cross‑border compute or share model weights internationally, seek specialized trade and compliance counsel.
• Where can researchers get compute access? Explore the NAIRR pilot and other public programs; pair access with strong data governance and security practices.

The bottom line

“AI in the United States” describes an active, adaptive system: private innovation and infrastructure, public standards and procurement, and a research base that continuously feeds new capabilities. For operators, clarity comes from building a repeatable AI risk program, aligning to recognized standards, and tracking the evolving federal–state mosaic. For policymakers, the challenge is keeping incentives for innovation aligned with protections for safety, civil rights, and national security—and doing so in ways organizations can implement.