- within Technology topic(s)
- in United States
- within Law Department Performance topic(s)
A FEDERAL AGENCY ROAD MAP FOR NEXT-GENERATION AI INFRASTRUCTURE
Executive Summary: Federal agencies must act decisively to maintain America's AI leadership. This paper provides agency leaders with a practical framework for establishing AI programs that balance innovation with security, speed with compliance, and transformation with risk management.
The recommendations herein are based on current federal mandates, emerging best practices, and lessons learned from early agency implementations.
1. STRATEGIC CONTEX
Federal Agencies Face an AI Imperative
Your agency operates at a critical inflection point. The rapid advancement of artificial intelligence presents both unprecedented opportunities and complex challenges for federal operations.1 While private sector adoption accelerates, government agencies must navigate unique requirements: statutory obligations, security imperatives, and public accountability standards that commercial entities do not face.2
Current federal policy establishes clear expectations: Agencies must integrate AI capabilities while maintaining robust governance frameworks.3 The challenge is not whether to adopt AI, but how to do so responsibly and effectively within existing operational constraints.
What Agency Leaders Need to Know About AI Models
For decision-making purposes, agency leadership should understand AI models as sophisticated analytical tools that process data to support mission objectives. Unlike traditional software that follows predetermined rules, AI models learn patterns from data to generate insights, predictions, or recommendations.4
Defining AI Models in the Federal Context
An AI model, as defined by 15 U.S.C. § 9401(3), is "a machinebased system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments."5 This definition distinguishes AI from basic automation, standard calculations, or simple "if-then" rule-based systems. AI models introduce learning capabilities that enable pattern recognition, adaptation, and improved performance over time through exposure to data.
The Fundamental Shift in Computing Paradigm
Traditional computing systems operate on explicit instructions— if X, then Y. Every outcome is predetermined by programmers. AI models represent a paradigm shift: They learn from examples to identify patterns humans might never recognize.6 This capability enables agencies to:
- Process unstructured data: Convert millions of pages of text, images, or audio into actionable intelligence
- Identify hidden patterns: Detect fraud, security threats, or system failures before they manifest
- Scale human expertise: Apply specialist knowledge across millions of cases simultaneously
- Predict future states: Forecast resource needs, maintenance requirements, or emerging risks
Types of AI Models Relevant to Federal Operations
Different agency missions require different AI approaches. Understanding these categories helps leaders make informed investment decisions:7
1 Foundation Models
Large, general-purpose models trained on vast datasets (such as GPT-4, Claude, and Gemini): These models can be adapted for multiple tasks without retraining. Agencies use foundation models for document analysis, report generation, and citizen services. Cost: $10 million–$100 million to develop, $1,000– $10,000 per month to operate.
2 Sovereign Models
AI models developed and controlled entirely within U.S. government infrastructure, addressing national security requirements and reducing dependency on foreign-controlled commercial systems.8 These models operate within federal security boundaries, ensuring data sovereignty and compliance with classification requirements. Critical for defense, intelligence, and sensitive civilian applications where foreign influence or data exfiltration risks are unacceptable.
3 Commercial Models
Commercially developed AI solutions available through vendor licensing or cloud services. These offer rapid deployment and proven capabilities but require careful evaluation of data handling practices, security controls, and potential dependencies on foreign infrastructure or ownership.9 Federal procurement should prioritize vendors with FedRAMP authorization and transparent supply chain documentation.
4 Specialized Models
Purpose-built models for specific tasks (image recognition, language translation, anomaly detection): These offer superior performance for narrow applications. Example: TSA's threat detection models process millions of x-ray images daily. Cost: $100,000–$5 million to develop, $100–$1,000 per month to operate.
5 Fine-Tuned Models
Foundation models adapted with agency-specific data: These combine broad capabilities with domain expertise. Example: VA's medical diagnosis assistant trained on veteran health records. Cost: $10,000–$500,000 to develop, similar operating costs to foundation models.
6 Agency-Specific Models
Custom models designed and trained specifically for an agency's unique mission requirements and operational environment.10 These provide maximum control over model behavior, security, and compliance but require significant investment in data infrastructure, technical talent, and ongoing maintenance. Ideal for agencies with highly specialized requirements that commercial or shared federal solutions cannot address.
7 Edge Model
Lightweight models that run on local devices without cloud connectivity: Critical for classified environments or field operations. Example: DoD's offline translation devices for deployed personnel. Cost: $50,000–$500,000 to develop, minimal operating costs.
To view the full article, click here.
Footnotes
1 National Security Commission on Artificial Intelligence, The Final Report (Washington, DC: National Security Commission on Artificial Intelligence, 2021).
2 U.S. Government Accountability Office, Artificial Intelligence: Agencies Have Begun Implementation but Need to Complete Key Requirements, GAO-24-105980 (Washington, DC: U.S. Government Accountability Office, 2023).
3 Executive Office of the President, "Removing Barriers to American Leadership in Artificial Intelligence," Executive Order 14179, The White House, January 23, 2025.
4 National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (Gaithersburg, MD: National Institute of Standards and Technology, 2023).
5 Executive Office of the President, "Removing Barriers to American Leadership in Artificial Intelligence."
6 National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework.
7 U.S. Department of Defense, DoD Data, Analytics, and Artificial Intelligence Adoption Strategy (Washington, DC: U.S. Department of Defense, November 2, 2023).
8 National Security Commission on Artificial Intelligence, The Final Report.
9 U.S. Government Accountability Office, Artificial Intelligence: An Accountability Framework for Federal Agencies and Other Entities, GAO-21-519SP (Washington, DC: U.S. Government Accountability Office, 2021).
10 U.S. Department of Defense, DoD Data, Analytics, and Artificial Intelligence Adoption Strategy.
Originally published 17 February, 2026
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
[View Source]
