Introduction
Artificial Intelligence is on the path to reshape how government functions, from the administration of public services to the back-end systems that keep government running.
This assessment evaluates the field of experimentation with and implementation of AI in state governments, and is an evolution of Code for America’s 2025 Government AI Landscape Assessment. In the last year, our team has worked to map new dimensions, and updated state evaluations reflect areas in which states have made progress and still have room to grow. This report also evolved to include a benefits access lens. However, many states experienced general AI rollouts that were not unique to benefits delivery and the administration of public services. This analysis was compiled primarily through public data. The full methodology is at the bottom of this report.
States are at varying stages of AI adoption—it’s not a single procurement decision or a one-time technology upgrade. It’s an institutional journey across agencies and departments. Most states are still early in this journey. Some are building governance frameworks. Others are piloting generative AI tools. A few are beginning to scale implementation. Very few have fully embedded AI into core operations with robust measurement and continuous improvement.
The AI journey can be understood as a progression across four stages:
- Readiness builds the foundation
- Piloting demonstrates what’s possible
- Implementation delivers results
- Impact ensures accountability and improvement
Each stage builds upon the previous one. With that in mind, this assessment focuses most on generative AI adoption, use, and value, while also picking up signals on previous implementations of predictive AI and intelligent automation, and new work in agentic AI. This site represents a short summary of our extensive research.
Download the PDF of the full 2026 research report
Reflecting a moment in time
This is a rapidly changing landscape and our assessment reflects a specific moment in time. Research culminated in March 2026. States are making changes, implementing new programs, and piloting new tools that may not be reflected in this assessment.
AI implementation in government
Read more about Code for America's vision for the future of government in an AI-powered world.
Stages of the journey
While these stages build on each other, they are distinct. Each of them has room for continuous growth over time, and each stage will continue to change as AI grows in maturity. Ideally, states will measure each of these phases and use agile feedback loops to advance them over time.
Stages
foundation Readiness
builds the foundation
Before AI can be used responsibly at scale, the right institutional conditions must exist. Foundational readiness focuses on three core pillars:
- Leadership evaluates the organizational structure and leadership dedicated to AI initiatives within state government
- Capacity evaluates the state's investments in developing AI literacy, skills, and expertise across its workforce
- Infrastructure evaluates the technical foundation—things like data accessibility, computing resources and platforms, and partnerships with technical vendors and service providers
experiment Piloting
demonstrates what’s possible
Once foundational elements are in place, governments begin structured experimentation and pilots. In this stage, we see states building:
- AI innovation labs within agencies or state-wide
- AI sandboxes for testing safely
- Pilot projects
- Limited deployments with clear guardrails and timeframes for evaluation
Such experimentation can reveal operational challenges like data or infrastructure limitations, ethical risks, workforce gaps, and procurement bottlenecks due to lack of established criteria or subject matter expertise.
build_circle Implementation
delivers results
At this stage, AI becomes embedded in government operations and systems. It moves from isolated pilots to scaled systems. Implementation may include:
- AI-assisted case management
- Public-facing chat assistants
- Predictive tools for benefits administration
- Fraud detection models
- Document automation
AI is no longer experimental—it’s operational. This introduces complexity that can require new capabilities: ongoing model monitoring, bias mitigation, cybersecurity concerns, and change management in a space where there are few established protocols.
insights Impact
ensures accountability and improvement
The final stage focuses on accountability and adaptation:
- AI systems must be monitored and measured, with ongoing evaluation of efficiency gains, cost savings, service quality improvements, and public trust
- Feedback loops are embedded
- Governance frameworks are updated based on lessons learned
- Training programs evolve alongside new technologies
This stage transforms AI from a set of technology projects into an agile ecosystem where leaders learn from and incorporate feedback while adapting to evolving AI. Learning gets brought forward into the next stage of AI use and adoption.
Assessment levels
Each stage of the journey is assessed using the following rubric.
Early
Initial steps with only basic foundational elements emerging
Developing
Core components in place with growing capabilities and some formalization
Established
Mature implementations with systematic approaches and demonstrated effectiveness
Advanced
Sophisticated, comprehensive frameworks and innovative, state-of-the-art approaches
States may find themselves at different maturity levels across different stages, reflecting their unique strengths and focus areas. This variation is expected and can help identify where to concentrate resources for advancement.
Trends & highlights
Across the states that are leading the way in AI implementation, there are some key lessons.
Leadership matters
States with strong executive leadership and dedicated AI governance bodies move faster.
Data infrastructure determines scalability
States with enterprise data platforms progress more quickly to operational AI.
Structured experimentation accelerates learning
Formal AI sandboxes and pilot programs generate better evidence.
Workforce readiness is essential
Training public employees in AI tools is a major driver of adoption.
The risk tolerance for government is lower
For government, there’s more at stake in responsibly moving from piloting to implementation than there is for industry. Government technology has a real impact on people’s lives, like access to benefits.
Measurement remains the next frontier
Even leading states are still developing systematic methods to measure public value from AI.
foundation Readiness
Early
-- states
Developing
-- states
Established
-- states
Advanced
-- states
In the Readiness stage, the central question is not yet how to deploy AI at scale, but who is accountable for it, how it should be governed, and what capabilities must exist to guide its responsible use. Over the past two years, nearly every state has entered this foundational stage in some capacity. A defining feature is the emphasis on guardrails and capacity building. States have focused heavily on responsible AI principles, ethical frameworks, transparency considerations, and risk mitigation. Many investments in readiness are happening at the state level, versus the agency level.
Spotlight
Georgia
Georgia’s strong leadership includes a chief data and AI officer who is part of the Georgia Technology Authority leadership team. Georgia invested in the infrastructure to enable AI, standing up a digital and physical AI Innovation Lab, a controlled sandbox environment where a roster of pre-vetted vendors can work directly with state agencies to conceive and deliver AI pilot projects.
Source: StateScoop
Spotlight
New York
New York also has a chief AI officer leadership position and launched the Empire AI initiative, which focuses on employee upskilling and building out infrastructure for AI testing and deployment. The initiative includes a proprietary generative AI sandbox environment (built by the Office of IT Services) where employees can practice using AI on work tasks with state-provided safe data, prevent data leaks and log learning outcomes.
Source: StateScoop
experiment Piloting
Early
-- states
Developing
-- states
Established
-- states
Advanced
-- states
In the Piloting stage, states move from policy conversations and training to hands-on experimentation. Nearly every state has engaged in some form of AI pilot, and many states started by experimenting with internal productivity tools. Staff used AI to summarize documents, draft communications, generate policy analyses, and support administrative workflows—use cases that were often framed as tools that help staff do their job more effectively.
In many cases, this experimentation is decentralized. Individual agencies explore use cases, often prompted by immediate workflow challenges or vendor outreach. While this approach generates energy and creativity, it can also produce uneven documentation, inconsistent risk review, and fragmented learning.
Spotlight
North Carolina
North Carolina’s Government Data Analytics Center (GDAC) had already piloted traditional AI and machine learning before the generative AI wave. The GDAC used this technology in areas like fraud detection to flag fraudulent benefit claims or tax evasion patterns, and opioid crisis analytics to predictively identify counties at risk of opioid overdose spikes. The GDAC now supports GenAI pilots like document summarization for policy briefs for the legislature’s research staff.
Source: Government Technology
Spotlight
Connecticut
Connecticut established an internal AI Enablement Lab within its IT division, providing a space for agencies to safely pilot AI use cases with privacy protections. Around 20 AI use cases were already deployed or in pilot by 2025, including trials of ChatGPT and Microsoft’s Copilot for office productivity. The state piloted AI in areas like document summarization in child services, automated Q&A for citizen inquiries, and predictive analytics in health care.
Source: NASCIO
build_circle Implementation
Early
-- states
Developing
-- states
Established
-- states
Advanced
-- states
The Implementation stage marks the transition from curiosity to commitment—taking successful AI pilots into the operational fabric of government. Most states are now somewhere within this operational transition, with many prioritizing efficiency gains over full service redesign.
The first operational AI systems tend to address backlog reduction, document processing, fraud detection, eligibility triage, and internal workflow automation. These use cases offer relatively clear return on investment and can be implemented without fundamentally altering public-facing systems. More ambitious transformations—such as AI-enabled case management systems, personalized benefits navigation, or automated adjudication support—are emerging more slowly. States with consolidated data platforms, cloud-first strategies, centralized IT governance, and strong data interoperability have progressed more quickly into Established or Advanced operational maturity.
Spotlight
Maryland
Maryland partnered with AI providers, including Anthropic, to deploy a generative AI-powered benefits navigation and eligibility guidance agent to help residents navigate programs such as food assistance, Medicaid, and housing support. The system answers questions about public benefits programs, helps residents identify which programs they qualify for, guides them through the application, and gives personalized information for household circumstances.
Source: StateScoop
Spotlight
Pennsylvania
Pennsylvania pioneered an Intelligent Document Processing service—where documents are scanned for legibility when users upload required documentation to the COMPASS benefits application system—to verify information in their application or renewal. This AI-driven tool screens for blurriness, image quality, and relevance to assist users in submitting accurate and readable information and help caseworkers quickly process information. These flags allow clients to resubmit unclear scans of necessary documents immediately, decreasing instances of illegible or incorrect documents by 80 percent and saving County Assistance Offices staff more than 700 hours of work. This work is now scaling up within the commonwealth.
Source: PA.gov
insights Impact
Early
-- states
Developing
-- states
Established
-- states
Advanced
-- states
The Impact stage represents the shift from implementation to impact, defined by whether the state has built the structures, culture, and discipline necessary to measure outcomes, learn from results, and adapt over time. The majority of states are still focused on defining guardrails rather than building longitudinal evaluation mechanisms. The result is a governance-heavy, measurement-light ecosystem in many jurisdictions.
There is very limited reporting on the impact of AI, and a truly continuous learning infrastructure remains concentrated in a small group of states. States that demonstrate higher maturity in this stage frequently rely on recurring automated decision system inventories or AI system registries to create visibility into where AI is being used and establish recurring reporting requirements.
Spotlight
Vermont
Vermont implemented a virtual assistant called ChatVT on its state portal to handle common questions about state programs. This AI chatbot has successfully managed thousands of citizen queries ranging from health services to DMV information. By providing 24/7 responses in plain language, the chatbot has improved access to information and relieved human call centers from a flood of routine inquiries. The state maintains an Automated Decision Systems Inventory requiring annual updates and performance review. Public reporting and advisory council documentation create recurring transparency and structured evaluation loops.
Source: VT Digger
Spotlight
Utah
Utah launched the nation’s first Office of Artificial Intelligence Policy (OAIP) and its AI Learning Lab program, explicitly designed to let AI innovations be piloted under close state oversight. In 2025, Utah approved a pilot with a local startup (Doctronic) to use an AI system for autonomously renewing routine prescriptions for chronic conditions. This pilot—the first state-sanctioned AI involvement in medical decision-making—operates in a sandbox with strict monitoring: OAIP evaluates safety protocols, patient outcomes, and ensures a physician is in the loop for exceptions.
Source: StateScoop
The Road Ahead
AI in State Government
Across the country, states are not waiting on the sidelines of this technological shift. They are stepping forward with urgency and a deep commitment to getting it right. We are seeing several states move from AI readiness to piloting and implementation.
The opportunity in front of us is not just about adopting new technology, but about shaping it in ways that are human-centered and grounded in real outcomes for communities.
When states lead with that mindset, they will do more than keep pace with innovation. They will define the future of public service in the AI era.”
State governments are poised for significant advancements in AI readiness over the next year, influenced by both industry developments and internal policy shifts. Here are some of the driving trends, and our predictions for the next year.
- As states implement agentic AI tools in the near future, the baselines for readiness could shift dramatically, impacting subsequent stages of this rubric.
- More states will formalize governance structures, launch broader and often mandatory workforce training initiatives, and expand sandbox and testing environments.
- We’ll see states move beyond isolated pilots toward coordinated experimentation portfolios. We will likely see a higher volume of successful pilots transition into operational programs, while agencies refine procurement strategies, technical standards, and governance practices based on lessons learned.
- We are likely to see more states formalize enterprise AI platforms, integrate AI into legacy modernization initiatives, and begin demanding stronger performance metrics.
- States that scale their AI implementations will be better able to articulate the impact of AI through clear evaluation frameworks that track efficiency gains, service improvements, and public value created by AI systems.
- Leading government agencies will start institutionalizing feedback loops that inform procurement, system design, and policy adjustments, enabling a cycle of continuous learning and responsible AI optimization.
Want to dive into the data with more depth?
Download the PDF of the full research report
Partner with us
Let's build your state's capacity to implement AI
We’ve partnered with states across the country with a solutions-oriented lens to technology. We take an evidence-based approach to problems and test relentlessly to deliver for both clients and state agencies.
Methodology
To develop a comprehensive view of state-level AI readiness and use, we conducted extensive desk research using publicly available materials with feedback loops from states and AI leaders. This included:
- Executive orders: Gubernatorial executive actions that established task forces, governance frameworks, or AI strategies.
- Legislation and policies: Laws and bills related to artificial intelligence.
- Agency guidance and reports: Strategic plans, policy documents, and technical guidance issued by state agencies, particularly IT departments.
- Media and trade articles: Local and national news coverage, civic tech blogs, and industry reporting.
- Direct state input: Opportunity for direct feedback and correction from states upon reviewing draft analysis.
- Advisory council: Review of all framing and the rubric by an advisory council.
Acknowledgements
The Government AI Landscape Assessment was made possible by the work of many people. We are especially grateful to Stephen Rockwell for leading our research and analysis efforts.
The Assessment was reviewed by our AI Advisory Council: Alicia Rouault, Deborah Jordan, Nishant Shah, Robert Lauwers, and Yuri Kim. The findings and conclusions contained within the Government AI Landscape Assessment are those of the authors and do not necessarily reflect positions or policies of the advisory council participants.
Download the full research report
There’s more to explore! Enter your email below to receive the full research report PDF for deeper insights and to review our methodology.
After downloading, you may receive additional communication from Code for America about other news that could be of interest to you.