The Numbers Behind the Gap
The 2026 NASCIO State CIO Survey data tells a coherent story when read together. Ninety-eight percent of states have enterprise policies and procedures on AI development and use. Eighty-two percent have GenAI projects in production — a question NASCIO asked for the first time this year, so there is no prior-year baseline. Thirty-five percent report enterprise-scale projects spanning the executive branch. Twenty-nine percent are running agentic AI in operations, with 35 percent planning to do so. Sixty-four percent of CIOs identify agentic AI as the emerging technology most likely to have the greatest impact in the next two to three years.
Each of those numbers signals deployment maturity. The 28 percent dedicated funding figure sits in direct contrast to all of them.
One CIO's comment in the survey puts the structural problem plainly: "We still do not have dedicated funding for AI which makes it difficult to move quickly with additional pilot projects, proofs of concepts, and expanding the sandbox." Given how widespread AI already is, NASCIO infers that funding "must be coming from other nondedicated sources and/or be more limited than states would like." Our reading is that the same constraint affecting new pilot capacity now also bears on the AI already running in production.
A fair caveat: the survey represents 51 self-reported responses from state and territory CIOs, including territories. The numbers describe posture, not audited outcomes.
What Running Production AI Without Dedicated Infrastructure Actually Looks Like
Production AI without its own program infrastructure does not fail dramatically. It drifts. The specific failure modes are predictable from the resource mismatch.
Security review gaps. Enterprise AI policy describes what AI must be reviewed for. Dedicated AI security staff and tooling do the reviewing. When only the policy exists, model updates, new integrations, and vendor-side changes go into production without the monitoring the policy assumed. The pattern resembles the shadow-IT period before formal cloud governance matured.
Vendor contract drift. AI capabilities embedded in enterprise software contracts are updated on vendor timelines. Without staff assigned to track those changes against state policy, agencies find out about capability additions or data handling changes after they are live. The 2026 NASCIO survey shows that 61 percent of states now report procurement terms and contract provisions for GenAI, and 49 percent require disclosure by software providers — but those provisions exist without necessarily having a person responsible for watching them.
Data governance theater. State CIOs named data readiness, governance, privacy, and sovereignty among the themes they expect to focus on over the next few years. The gap in that effort is specific: 98 percent of states have enterprise AI policies, but only 57 percent have implemented data governance covering data sources, quality, bias, and privacy. One CIO cautioned that their state's efforts in these areas "are very early, immature and/or not strictly enforced. We still lack broad understanding and socialization of our AI policy, making enforcement and governance difficult." Policy documents without governance staff produce the appearance of readiness without the operational substance.
Procurement without assessment. Agencies across the executive branch are evaluating and acquiring AI-enabled tools. Without centralized AI acquisition support — procurement officers trained in AI-specific contract language, standard evaluation criteria, and vendor disclosure requirements — acquisitions proceed against inconsistent standards. The result is a patchwork of AI commitments the enterprise has not evaluated as a whole.
The Workforce Cliff Compounds Everything
One CIO in the NASCIO survey described the state CIO organization as facing a "workforce cliff": "The people who hold the institutional knowledge of our legacy systems are retiring, the pipeline behind them is thin and AI is simultaneously redefining what an IT job even is."
Three problems arrive at once. The retirement wave drains institutional knowledge of legacy systems — COBOL-era mainframes, proprietary benefit platforms, decades-old integration architectures — that is not easily documented or transferred. AI is creating new job functions — AI risk analyst, model governance coordinator, AI acquisition specialist — that traditional civil service classification processes may be slow to accommodate. NASCIO's data offers a partial counter to that last point: about half of states have already hired someone with "AI" in their job title. The challenge may be less about the existence of the classification than about whether those hires are reaching the agencies that most need governance support.
Sixty-eight percent of CIOs in the survey expect more turbulence and disruption in their role over the next five years. The AI governance infrastructure being built now will be tested under conditions of institutional churn, not stability.
The Incoming Leadership Transition Problem
The 2026 elections placed 36 gubernatorial seats on the ballot. NASCIO notes this will likely bring many new state CIOs in 2027. One CIO's advice to incoming successors in the survey: make clear that "enterprise standards, security, accessibility, architecture and fiscal discipline are not optional."
The operational risk is straightforward. An incoming CIO who inherits production AI without a funded program infrastructure will spend the first months reconstructing what is running, where, and against what policy commitments. The current cohort can reduce that transition risk by getting dedicated AI program infrastructure into the upcoming budget process, so that whatever comes next has something institutional to build on.
Five Focus Areas, Five Funding Requirements
State CIOs named five themes they expect to focus on over the next few years: scaling from pilots to operational production, improving governance and standards, addressing AI-specific cybersecurity risks, exploring agentic AI, and improving data readiness. Each has a specific infrastructure requirement that general IT budgets typically do not include.
Scaling from pilots to production requires an enterprise program management function — not project-by-project management, but enterprise-level standard-setting, vendor oversight, and incident response coordination that individual agencies cannot sustain alone.
Improving governance and standards requires people whose job is governance — analysts who maintain AI inventories, track policy compliance, run periodic audits, and advise agencies on specific deployment questions.
Addressing AI cybersecurity requires security staff with AI-specific training and tooling. AI systems introduce attack surfaces that traditional vulnerability management frameworks do not fully cover. NIST's adversarial machine learning taxonomy (NIST AI 100-2e2025, March 2025) catalogs attack classes specific to AI systems, including evasion, data and model poisoning, model extraction, and direct and indirect prompt injection.
Exploring agentic AI responsibly requires a sandboxed environment with clear authorization boundaries and human accountability for every consequential action. NASCIO records one CIO's approach: "We will prove it internally, with a human accountable for every consequential action, before extending it to citizen-facing operations." Twenty-nine percent of states already use agentic AI in some state operation, mostly internal workflow, IT operations and contact-center work. For states still deciding how far to extend it, building oversight before deployment is easier than retrofitting it.
Improving data readiness requires data governance investments — cataloging, classification, lineage documentation — that are prerequisites for reliable AI, not just AI-enabling extras. The 41-percentage-point gap between states with AI policies (98%) and those with data governance implementation (57%, per the survey) is the clearest indicator of where this work stands.
What a Budget Ask Should Actually Contain
For state CIOs entering appropriations discussions — in most states for FY2028, though fiscal years and biennial cycles vary — the NASCIO data provides a defensible foundation for a dedicated AI program funding request. The case is not "AI is exciting." It is that production software running in state operations requires production program management, and state AI is production software.
A credible ask includes: an AI program office function (even if embedded in the existing CIO organization), AI security tooling and training for the security operations team, AI acquisition support capacity, data governance investment tied to AI-enabling outcomes, and a sandbox/testing environment for evaluating agentic AI before operational deployment.
A legitimate alternative exists for states with limited AI footprints: a shared-service AI program or enterprise chargeback model, where agencies that benefit contribute to a shared governance pool. That approach works when usage is bounded and concentrated; it starts to break down as GenAI embeds across enterprise platforms and governance work disperses. Many CIOs in the survey call the chargeback model inflexible and want sustainable enterprise funding instead — treat it as a bridge rather than an end state. Know which situation you are in before defaulting to it.
The choice is not whether to pay for governance. It is whether to pay now, through a planned program, or later, when incidents, contract exposures, or legislative inquiries force it.
Decision Questions for State CIOs
Before entering budget season, these questions identify the most consequential gaps:
- For each production AI system running today, who is responsible for security review when the vendor updates the model or the integration?
- Which agency AI acquisitions have been evaluated against the enterprise AI policy, and by whom?
- If the state's AI inventory were audited tomorrow, what percentage of vendor-embedded AI features in enterprise contracts would appear in it?
- For states using AI in consequential decisions — benefits eligibility, case routing, fraud detection — what is the documented human review process, and who verifies it is followed? Colorado's SB26-189 (signed May 14, 2026; repeals and replaces SB24-205; requirements apply to decisions on or after January 1, 2027) requires notice, explanation of adverse outcomes, and meaningful human review when automated decision-making technology materially influences a consequential decision. Enforcement awaits Attorney General rulemaking. Jones Walker's post-election analysis advises organizations to plan for state obligations rather than federal relief. Whether your state uses similar technology against a comparable standard is worth confirming.
- What does the FY2028 budget request say about AI, specifically? If the answer is "nothing, it is covered in existing lines," that is an answer about risk tolerance, not funding adequacy.
Sources and further reading
- NASCIO, The 2026 State CIO Survey: The State CIO in Motion — Priorities, Pressures and Progress (51 state and territory CIOs, fielded summer 2026)
- NASCIO, 2026 State CIO Top 10 Priorities (released December 2025)
- Colorado General Assembly, SB26-189, Automated Decision-Making Technology, signed May 14, 2026
- Jones Walker LLP, AI After the Midterms, November 6, 2026
- NIST, AI 100-2e2025, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations, March 2025
Spartan X's AI advisory work with state leadership in Colorado and Florida, along with the state platforms it has designed and delivered, sits in this gap between adoption and operational readiness, where governance and risk frameworks decide whether AI becomes something a state can operate.



