From the State Government edition of October 22, 2026
InnovateUS Observatory's first report: 51% of public servants in its basic AI course had never tried generative AI, and uncertainty about relevance, skills and permission, not resistance, drives non-use
InnovateUS Observatory of Public Sector AI (Burnes Center for Social Change, Northeastern University), with Rutgers and Elon University researchers · State and public-sector AI workforce readiness and training · United States (all 50 states and DC; state, local, tribal, territorial and federal learners)
- Publisher
- Reboot Democracy (InnovateUS / Burnes Center)
- Original publication
- 2026-10-20
- Source retrieved
- 2026-10-23
What happened
The Observatory's first report analyzes more than 106,000 pre-course and 67,000 post-course surveys from InnovateUS's introductory 'Using Generative AI at Work' course (June 2024 to May 2026), with 22,303 matched learners. Among course takers, 51% had never tried generative AI and another 14% had tried it only outside work; 64% of public safety learners had never tried it versus 23% of digital services staff. Among non-users, 28% said AI does not apply to their work, 23% did not know how to use it and 19% did not know whether they were allowed (the report's overall figures); among managers, 24% cited not knowing how and 22% not knowing whether it was permitted. Self-reported 'a lot of knowledge' rose from 13% to 45% across pre- and post-course samples, and very confident learners planned 3.8 AI tasks versus 1.5 for those not confident. In Washington, DC's mandatory training (32,981 enrolled, about 86% of staff), preliminary results show 41% had never tried generative AI.
Why it matters
Why now: released Oct 20 and cited in Oct 21 reporting on slow state uptake, it is the largest dataset yet on whether public employees given AI tools are ready to use them. It reframes state AI rollouts: license counts overstate adoption when half of learners have never tried the tools and many do not know what is allowed.
Evidence and measured results
Opt-in surveys of course takers, not a representative sample; the authors say margins of error cannot be calculated. Pre/post knowledge and confidence figures mostly compare different respondent groups (22,303 matched). Measures are self-reported knowledge, confidence and intended use immediately after the course, with no follow-up on workplace behavior or resident outcomes. Voluntary learners start more knowledgeable and optimistic, so familiarity may be overstated. The research is produced by the training provider evaluating its own learners.
Limitations and uncertainty
Provider-run research on its own learners; opt-in and self-selected; self-reported; cross-sectional pre/post comparisons; mixes state, local, federal and other public workers; DC results preliminary. Not evidence that training improves services.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-22; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- States buy AI tools, but many employees have never used generative AI, do not see its relevance or do not know whether they are allowed to use it.
- Stakeholders
- State chief AI officer
- State HR or workforce development director
- Agency training leads
- Agency managers and supervisors
- CISO or privacy officer
- Discovery questions
- What share of licensed staff actively use your AI tools each month?
- Do employees know which AI tools and uses are permitted?
- Have managers been trained before their staff?
- Which roles have the lowest uptake?
- How do you measure whether training changed work?
- Value hypothesis
- Clear permissions, manager training and role-specific examples may raise responsible use among licensed staff; this report shows confidence gains after training, not workplace outcomes.
- Potential engagement
- Ten-week AI readiness sprint for one agency: baseline usage telemetry and a staff survey, publish permitted-use guidance, train managers, run two role-specific workshops, and remeasure.
- Unsupported claims
- Training improves services or productivity
- Findings represent the whole public workforce
- A specific national weekly-use rate
- Causal link between confidence and use
- Lasting behavior change
Pre-sales engineering
Role takeaway
- Fit
- States or agencies with enterprise AI licenses deployed broadly but uneven, unmeasured adoption across roles.
- Architecture and integration
- Aggregate usage telemetry from the enterprise AI tool joined to learning-management completion records and role codes for reporting.
- Prerequisites
- Enterprise AI tool with admin usage reporting
- Learning management system data
- Role or job-class mapping
- Privacy review of usage analytics
- Constraints
- Self-reported survey data only in this source
- Union and privacy limits on individual monitoring
- Role mixes vary by jurisdiction
- Multi-level sample (not state-only)
- Security
- Make the approved AI environment the default and block or warn on unapproved public tools, since uncertainty about permitted tools can push staff to consumer AI; report usage only in aggregate.
- Proof of value
- Compare active-use rates for trained versus untrained staff
- Check usage by role against survey patterns
- Measure persistence 90 days after training
- Track help-desk questions on permitted use before and after guidance
Delivery
Role takeaway
- Work
- Design and run a workforce AI readiness program tied to an enterprise tool rollout, with permitted-use guidance and manager-first training.
- Dependencies
- Published acceptable-use policy
- HR and training office
- Agency managers
- Enterprise AI tool admin access
- Ownership
- State HR or workforce development office jointly with the chief AI officer.
- Skills and adoption
- Train managers first, then staff, using role-specific examples for low-uptake roles such as public safety, legal and administrative support.
- Governance checkpoints
- Permitted-use list approved and published
- Privacy review of usage reporting
- Manager training completion before staff launch
- Follow-up evaluation at 90 days
- Acceptance criteria
- Proposed: published permitted-use guidance within 30 days
- Proposed: at least 80% of managers trained before staff sessions
- Proposed: role-level active-use baseline and 90-day remeasurement reported
- Risks
- Training without tool access or permission
- Survey fatigue and self-selection
- Over-reading confidence as competence
- Monitoring concerns from staff and unions
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Limited relevance: not a technology study. Interpretation: usage data from enterprise AI tools would be needed to test the survey's self-reported findings, which the Observatory says it plans to pair with surveys.
Governance
Who approves, reviews and stays accountable for outcomes?
About one in five non-users do not know whether they may use AI, and the report argues the absence of a policy functions as a policy. States should publish plain-language permitted-use guidance and approved-tool lists and train managers, who report the same permission gaps.
Security and privacy
What data, permissions and controls need testing?
The blog cites an untrained Victoria, Australia child protection worker who pasted confidential details into ChatGPT, illustrating the risk when tools are available without training; the survey itself does not measure data incidents.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Readiness gaps are uneven by role (64% of public safety learners never tried generative AI versus 23% of digital services staff); mandatory training reached about 86% of DC staff.
Procurement
What should contracts, pricing and exit terms secure?
License purchases should be paired with training and measurement budgets, since access alone does not produce use.
Operating model
Which teams own the service once it runs?
The report recommends training the whole workforce, training managers early, stating permitted uses clearly, teaching AI in job context and measuring outcomes rather than uptake or satisfaction.
Publication history
- 2026-10-22State Government · Issue 454 resources
Stable resource ID: innovateus-observatory-public-sector-ai-first-report-51pct-never-tried-genai-20261020