Helping state and local government decide what AI is worth buying, and get real value from what they do buy.
Working across state and local governmentThe council brings together legislators, agency leadership, municipal staff and university researchers, with the Office of Science and Technology contributing to its published work. Members include the Georgia State Senate, the cities of Brookhaven and Alpharetta, the Atlanta BeltLine, Georgia Tech, DePaul University and KPMG.
Procuring artificial intelligence as a single monolithic technology is a reliable way to spend public funds badly and damage public confidence at the same time.
This council breaks that word into categories a state or local government can actually make decisions about, and works out what it takes to get value from each of them. Members are lawmakers, agency leadership, municipal staff and university researchers, and the work is aimed at the people who have to sign the contract and live with what they bought.
The approach we argue for is human centred rather than technology led. Governments that begin by giving their own workforce better tools build the internal knowledge needed to judge larger systems later. Modernising procurement and breaking down departmental silos matters more in the early stages than any particular model does.
The council is a standing body, not a publishing project. Its work runs across convening, applied research, workforce, procurement and policy, and the six federal pillars in the consortium's policy posture shape all of it.
Legislators, agency leadership, municipal staff, university researchers and technology practitioners rarely sit at the same table. Getting them there, regularly and off the record, is the single most useful thing this council does. Most bad AI decisions in government come from one of these groups acting without the others.
The council places work inside operating municipalities rather than laboratories, so methods are tested against real budgets, real procurement rules and real public scrutiny. What one city settles about governance and trust should not have to be settled again by the next.
Extending the posture's education and workforce pillar into government. AI literacy across leadership and staff is what makes oversight genuine, and an agency that cannot evaluate a vendor claim ends up delegating the decision to the vendor.
Applying the posture's case for regulatory sandboxes and tiered, risk based treatment to how governments buy. The council works on specification, evaluation and contracting, and on running pilots that end in a decision rather than a demonstration.
Drawing on the posture's security and privacy pillar. Classification, role based access, secure sharing agreements between agencies, and resilience against the cyber risk that ageing public infrastructure carries. These need to be settled before deployment, not after.
Turning what members learn in operations into guidance lawmakers can use, through published recommendations, direct engagement with agencies, and mapping those recommendations to funding programmes so that good policy has a route to being paid for.
Committees and subcommittees focused on AI regulation keep the body out of red tape and regulatory stagnation. Sandboxes and testbeds allow controlled experimentation without undue burden.
Leaders translate legislative theory into executive practice across very different domains and workforce groups, which is where most adoption succeeds or stalls.
Subject matter experts turn technical change into guidance a policymaker can use, though agencies should weigh that guidance against operational reality and avoid vendor driven advice.
Residents are the beneficiaries and the judges of any of this. Transparency about where AI is used, especially in decisions affecting access to services, is a precondition rather than a nicety.
They convene the conversation usefully, provided procurement decisions stay grounded in transparent evaluation and measurable outcomes rather than commercial interest.
The council works to five steps in order, because most failed adoptions skip straight to the fifth.
Decide what outcome is being pursued before deciding what technology might deliver it.
Work out which of the five groups the decision touches, because each has different authority and different exposure.
Establish which category of tool the problem actually calls for, rather than treating AI as one purchasable thing.
Run the decision against data governance, strategy alignment and team composition before committing funds.
Move decisively once the first four are answered. Indecision has a cost as real as a bad procurement.
Governance frameworks for security, privacy and access belong in place before AI is deployed at scale rather than after. That means classifying data by sensitivity, setting access controls by role, and meeting the obligations that already govern public records, personal information and cybersecurity.
Systems should be assessed for model integrity, third party integration risk and unauthorised exposure. Secure data sharing agreements between agencies reduce duplicated effort, and investment in logging, monitoring and incident response lets adoption scale without spending public trust to do it.
Strategy means aligning investment with the mission and the service objectives of the organisation. Isolated pilots are the common failure mode. The better pattern is to find high value use cases that produce measurable outcomes while building capability other departments can reuse.
What tends to be present when this works: executive sponsorship, a governance committee, a workforce development plan, and standards for ethical use. Evaluation metrics should cover service delivery, efficiency, equity and risk together.
Adoption needs multidisciplinary teams rather than a technology function working alone. That means programme managers, data specialists, cybersecurity staff, procurement officials, legal advisors and frontline operational people in the same room.
Investment in AI literacy across leadership and staff is what makes oversight meaningful. Without it, decisions get delegated by default to whoever sounds most confident.
A national programme applying computer vision to pedestrian and cyclist movement on multi-use trails and public spaces, so that cities can plan from measured behaviour rather than assumption.
Launched in August 2024 with Georgia Institute of Technology, DePaul University and participating municipalities, the programme turns traffic patterns into planning insight. Cities use it to understand how public space is actually used, to support economic assessment and community development, and to identify where carbon emissions can realistically be reduced.
It is offered as a service rather than a product, covering hardware, software and ongoing support, which matters for municipalities that have neither the procurement appetite nor the staff to run infrastructure of this kind themselves.
Privacy is handled at the point of collection. Personally identifiable information is obscured at the edge, synthetic data modelling is used in place of real records where possible, and data security governs the whole pipeline. The programme reports how spaces are used, not who used them.
The City of Brookhaven and the Atlanta BeltLine were the first deployments, with Brookhaven serving as an anchor municipality where methods can be tested under real operating conditions. The programme is designed for reuse, so what is settled once about procurement, governance and public trust does not have to be settled again by the next city.
The current phase extends the work beyond counting. Classical computer vision handles perception while vision language models handle interpretation, which lets the system describe how a space is being used rather than only how many people passed through it. With Georgia Tech the programme also runs controlled experiments, changing something in the environment and measuring the response.
Data science and AI have the potential to transform urban living by providing critical insights for better decision making and sustainability. Ilyas UstunDirector of Data Science, DePaul University
By bringing together diverse expertise, we are not only advancing technology but also ensuring its positive impact on society and the environment. Michael BraunSmart City Lead, AI Applied Consortium
Breaking the word apart is the point. Each of these behaves differently, fails differently, and needs different things in place before it works.
Generative AI, predictive analytics and intelligent automation applied to productivity and decision making, which matters most where agencies are short staffed and carrying administrative load.
Optical character recognition and secure indexing built to handle sensitive government records without exposing them.
Conversational systems covering unemployment, tax support, transit information and emergency communication. More advanced agents complete transactions across systems while keeping audit trails and human escalation paths.
Routing and coordination across permit approvals, procurement review, inspection scheduling and case management, including intelligent dispatch for emergency services and automated eligibility determination for benefits.
Automated threat detection, anomaly identification and incident response. Machine learning reads network traffic and prioritises alerts faster than rule based systems, which matters given ageing infrastructure and ransomware exposure.
Cleaning, integration, cataloguing and lineage tracking. Applied well it improves budget forecasting, infrastructure maintenance planning, public health monitoring and fraud detection.
Traffic monitoring, infrastructure inspection, environmental analysis and asset inventory. Transportation agencies use it for roadway hazards, and emergency management for aerial disaster assessment. It needs edge computing in the field and explicit governance on privacy and responsible use of visual data.
Principles and policy for AI adoption in state and local government, published July 2026 by the Public Sector Advisory Group.
The brief sets out a sequence rather than a shopping list: define the goal, identify which parts of the ecosystem are involved, categorise the technology, apply the principles, then act. It covers data classification and access control, what an enterprise AI strategy should contain, and how to staff a programme team with the mix of technical, legal, procurement and frontline people it needs.
It extends the consortium's 2025 response to the federal AI Action Plan RFI into state and local government.
What members learn in their own operations only matters if it travels. The council runs a deliberate path from evidence to policy to funding, and the last stage feeds the first.
Industry specific recommendations and priorities, written by the councils that hold the expertise.
Briefs go back to industry to align priorities, gather insight and build consensus before anything is published.
Priorities and recommendations are communicated to inform policy and programmes.
Recommendations are mapped against federal funding programmes and initiatives.
Members use the briefs to strengthen proposals and drive investment into AI work.
Continuous impact: improving competitiveness, driving innovation and advancing responsible AI adoption.
The people who will inherit these systems are rarely part of the decisions that shape them. This programme changes that.
Structured discussions at universities, currently the University of Kentucky and DePaul.
Themes and insights are synthesised into findings that can actually be acted on.
Student perspectives are integrated into testimony, data and worked examples.
Findings support federal initiatives, policy development and funding priorities.
The next generation voice enters federal AI and workforce decisions.
The loop runs continuously, capturing, learning from and elevating student perspectives with each cycle.
Rebbeca Beers, Public Sector AI Lead at Salesforce, and Michael Braun, with Konrad Konarski.
Legislature, municipalities, universities and advisory firms, working together on research, pilots and policy across state and local government.
Research university
Research university
Municipality 
Civic infrastructure
Municipality
State legislature
Advisory
Contributing institutionLegislators, city officials, university researchers and public sector practitioners, each bringing a different kind of authority to the same problem.
Chair, Senate Finance Committee
Public Sector Advisor, AI Applied Consortium · Senior Advisor
Director of Strategic Partnerships
Professor and Department Head
Director of Analytics
Principal Engineer
Representative
Membership suits people making AI decisions inside state or local government, along with researchers working on urban systems who want their work tested in a real city.
Members join quarterly working sessions and take part in the council's research and policy work. In practice this comes to about a day each quarter.
You get peers facing the same procurement and governance questions, early access to what the pilots produce, and a hand in the guidance the council puts in front of lawmakers.
We ask members to share what adoption actually cost them, including the parts that went badly. Other cities can only learn from accurate accounts.
[CONFIRM — state dues plainly, or “no fee” if that is the model.]
