Stay Updated

Public Sector AI Trends 2026: Research Analysis

In 2026, government AI moved from hype to operations: adoption is near-universal, but fewer than one in five organizations have scaled it, and governance, workforce and measurement now decide who succeeds.

  • AI is the top priority. For the first time, state CIOs ranked AI above cybersecurity, and federal AI use cases more than doubled to 3,611.
  • Execution is the gap. 90% of states run AI pilots, but only 20% of state and local IT leaders have scaled AI across multiple workflows.
  • Agentic AI has arrived. Roughly half of federal agencies are piloting or planning AI agents, and CISA issued its first agent-specific security guidance.
  • Platforms beat point tools. GSA’s OneGov/USAi and California’s Poppy show a shift to shared, multi-model, government-controlled platforms.
  • Regulation is contested. A federal executive order targets state AI laws while states passed 109 new AI laws in the first half of 2026.
  • People are the constraint. Only about a quarter of agencies have trained employees on AI.

The priority for leaders: fund fewer pilots with clear paths to scale, harden governance for high-impact and agentic AI, and make workforce readiness the critical path.

The 2026 landscape

Government AI in 2026 is defined by a federal push to accelerate adoption, a federal-state fight over who regulates, and fiscal pressure that demands measurable returns.

Federal: acceleration, centralized buying. OMB memos M-25-21 and M-25-22 (April 2025) require every agency to name a Chief AI Officer, publish an AI strategy, inventory use cases and apply minimum risk practices to “high-impact” AI. A December 2025 OMB memo added “unbiased AI” requirements for contracted LLMs, with agency policy updates due March 11, 2026.

GSA has become the federal AI buying hub. OneGov treats government as one customer; it has drawn more than two dozen vendors and identified about $1.18 billion in savings. Nearly 3.4 million users across government can now reach AI through it, and the USAi sandbox (launched August 2025) feeds usage data into the next round of negotiations.

Federal vs. state: the preemption fight. Executive Order 14365 (December 11, 2025) created a DOJ AI Litigation Task Force, formalized January 9, 2026, to challenge state AI laws. It also asked Commerce to flag “onerous” state laws and tied some broadband funding to state compliance. States kept legislating anyway: 109 AI-specific laws were enacted in 2026 through early July. Counsel broadly advise that an executive order cannot void a statute, so state laws remain in force while litigation plays out.

States: from pilots to production. California moved its GenAI pilots (CDTFA, Caltrans, CDPH, CalHHS) into deployment agreements for traffic, road safety and call-center service, using its RFI2 procurement method. The LAO warned the timeline lacked full scope and cost estimates.

International. EU AI Act obligations continue to phase in through 2026, and SAS expects regulators to begin first enforcement actions, pushing governance from compliance task to innovation driver.

The through-line of 2026 is a shift from experimentation to accountable operation: adoption is near-universal, but scale, measurement and governance lag behind.

# Trend Evidence What it means for agencies
1 Pilots to production, and the scaling gap 82% of state CIOs say staff use GenAI daily and 90% run pilots (GovTech/NASCIO), yet only 20% of state and local IT leaders have scaled AI across multiple workflows (MeriTalk) Funding and success criteria must be set before a pilot starts, or it stalls in “pilot paralysis” (APA)
2 AI becomes the top CIO priority For the first time in 20 years, AI displaced cybersecurity as the No. 1 NASCIO state CIO priority for 2026, with emphasis on governance, ethics, data quality and workforce AI moves from innovation office to enterprise portfolio, with budget and accountability to match
3 Agentic AI enters mission work 53% of federal agencies are exploring or planning agentic pilots and 15% are implementing (ServiceNow/Market Connections); 58% of federal IT leaders report deployed or piloted agents (Booz Allen). CISA issued its first agentic AI security guidance on May 7, 2026 Agents that act, not just answer, require new controls: permissions, audit trails, human checkpoints
4 Shared platforms replace one-off tools Federal: GSA’s OneGov and USAi. California: Poppy, a state-built, multi-model assistant that went statewide July 1, 2026 after 70+ departments piloted it Central platforms lower cost and risk, avoid vendor lock-in, and generate usage data for smarter buying
5 Use cases multiply; high-impact AI grows Federal inventory hit 3,611 use cases across 56 agencies in 2025, up 105%; 445 were “high-impact,” 215 at VA alone Risk management must scale with volume; inventories become a live governance tool, not a compliance report
6 Regulatory fragmentation EO 14365 targets state AI laws while states enacted 109 AI laws in the first half of 2026 (SecurePrivacy) Agencies and vendors need dual-track compliance until courts settle preemption
7 Workforce readiness lags adoption Only 21% of local and state agencies actively use AI and just 24% have trained staff (NEOGOV, 4,200+ respondents) AI literacy becomes a core competency; change management decides whether tools are used
8 ROI and trust pressure Just 5% of integrated AI pilots returned meaningful value (GovTech); 79% of federal IT leaders fear AI-accelerated cyberattacks (Booz Allen) Every deployment needs a baseline, a public-value metric, and a security case

Where use is concentrated. Federal AI use clusters in a few large departments. HHS (447), NASA (425), VA (367), Energy (340) and Justice (314) lead the 2025 inventory, and Brookings finds the 11 small agencies reporting submitted only 2% of the total.

Maturity is a staircase, not a switch. Code for America’s 2026 assessment finds nearly every state has piloted AI, a small group has scaled across agencies, and very few have embedded it with continuous measurement. A state mature in GenAI may still be at the readiness stage for agentic AI.

Governance, risk and workforce implications

Governance in 2026 is less about whether to use AI and more about controlling what it does at scale. Four pressures stand out.

1. Governance frameworks exist, but enforcement does not. The ServiceNow-sponsored federal survey found a wide gap between how important leaders rate governance and how much they have implemented. Only 20% of public sector organizations report mature data governance, per NASCIO data cited by NTT DATA. Without clean, governed data, AI outputs cannot be trusted at scale.

2. High-impact AI needs a defined control set. M-25-21 requires risk assessment, testing and ongoing monitoring for AI that drives decisions affecting rights or safety. The share of federal use cases flagged high-impact fell from 16.5% to 12.3%, but the count rose from 351 to 445 (Brookings). Brookings also notes 32 cases were downgraded between years, which warrants scrutiny of classification practices.

3. Agentic AI shifts risk from content to action. CISA and international partners’ May 2026 guidance on careful agentic adoption is the first federal document aimed at autonomous agents. Top federal concerns are protecting sensitive data (56%) and adversaries using AI to speed attacks (79% very or extremely concerned), per Booz Allen.

4. Control over models and vendors is contested. OMB’s December 2025 memo requires contracted LLMs to meet “unbiased AI” principles. The White House is also weighing draft language asserting government authority over how purchased AI is used, limiting vendor-imposed restrictions. Procurement terms now carry policy weight.

Workforce. Adoption depends on people more than platforms. California has built 20+ state AI trainings, a five-course certificate and boot camps, with UC Davis, Sacramento State and Cisco as partners (PSHRA). California’s approach also pairs deployment with union engagement rather than headcount cuts (Governor’s Office). Nationally, though, only about a quarter of agencies have trained staff, so change management, role redesign and trust-building remain the binding constraint.

Implications for California and state agencies

California is both the nation’s most active AI regulator and one of its most advanced state adopters, which puts its agencies at the center of the federal-state conflict.

A shared GenAI platform is now standard infrastructure. Poppy moved from pilot to statewide availability on July 1, 2026. It is multi-model, keeps data in a state-managed environment, detects PII automatically and bars model training on department data. CDT’s FY 2026-27 rate memo offers free access for up to 50 users per department, then $8 per user, and a budget change proposal requests $1 million to broaden adoption. Departments should default to Poppy before buying standalone tools.

The legal environment is layered. California’s 2026 AI rules mostly regulate developers and businesses, but they shape what agencies buy and how vendors behave.

Law or rule What it does Agency relevance
SB 53 (effective Jan. 1, 2026) Frontier developers publish safety frameworks and report critical incidents to Cal OES within 15 days Cal OES gains an incident-intake role; vendor frameworks become procurement evidence
CPPA ADMT rules (compliance by Jan. 1, 2027) Pre-use notices, opt-outs and risk assessments for automated significant decisions Contractors using ADMT on Californians’ data need compliant notices and assessments
AB 2013 / SB 942 Training-data disclosure; AI-content detection (effective Aug. 2, 2026) Supports transparency requirements in GenAI contracts

Federal preemption is a live risk. EO 14365 explicitly targets California-style rules and links some broadband dollars to state compliance. Agencies should keep complying with state law, document compliance, and track DOJ Task Force and Commerce actions.

Oversight expects evidence. The LAO has pressed for full scope and cost data on GenAI projects. Departments should expect budget asks to require baselines, benefit metrics and post-deployment results.

Labor partnership is a differentiator. California positions AI as augmenting state workers and protecting union jobs. That framing lowers resistance but raises the bar for transparent change management and retraining.

Recommendations for public sector leaders

Leaders who win in 2026 treat AI as an enterprise change program with measurable outcomes, not a technology rollout.

  1. Fund scale, not just pilots. Approve pilots only with a named owner, a baseline, a success threshold and a funded path to production.
  2. Standardize on shared platforms. Use enterprise platforms (OneGov, USAi, Poppy) for general GenAI; reserve custom builds for mission-specific needs.
  3. Make the AI inventory a management tool. Tie each use case to risk tier, owner, data sources and outcome metrics, and review high-impact cases quarterly.
  4. Write an agentic AI control standard now. Define permission scopes, human approval points, logging and rollback before agents touch case files or payments. Use CISA’s May 2026 guidance as the baseline.
  5. Invest in data governance first. With only 20% of organizations reporting mature data governance, data quality is the main limiter on trustworthy AI.
  6. Run dual-track compliance. Comply with state law as written while tracking federal preemption actions; document decisions for auditors and the legislature.
  7. Embed procurement guardrails. Require vendor safety frameworks, data-use limits, no-training clauses, audit rights and exit terms in every AI contract.
  8. Treat workforce readiness as the critical path. Pair every deployment with role-based training, manager coaching, union engagement and redesigned workflows.
  9. Report public value, not just savings. Measure faster service, fewer errors, better access (such as multilingual service) and risk avoided, alongside cost.

Outlook: what to watch into 2027

The next 12 months will decide whether 2026’s adoption turns into durable, measurable public value.

Watch item Why it matters Timing
Federal preemption litigation and any congressional AI framework Determines whether state AI laws survive as written Ongoing; years possible
GSA charging agencies for USAi Tests whether shared AI platforms sustain under cost recovery (MeriTalk) FY 2027
OneGov agreements expiring Pricing and vendor mix reset as short-term deals become long-term programs About a year out
CPPA ADMT compliance deadline Vendors’ automated decisions on Californians face notice and assessment rules Jan. 1, 2027
Poppy adoption and cost recovery First full year shows whether a state-built platform scales economically FY 2026-27
Agentic AI incidents A visible failure would reset risk appetite and oversight rules Any time
EU AI Act enforcement First actions set global expectations for public sector AI accountability 2026-2027

Sources

Download Resource

Get new insights, perspectives, and resources delivered to your inbox.