Georgia publishes local-government AI guide urging safeguards before purchases

Georgia’s cities and counties now have a state-developed starting point for using AI without handing over public judgment to it. The Georgia Technology Authority’s Office of Artificial Intelligence released its Local Government AI Playbook on November 12, 2026, recommending that local officials establish policies, oversight and employee training before purchasing AI products.

The guide targets communities without dedicated AI teams, formal policies or resources to build their own governance framework, according to StateScoop. Georgia chief digital and AI officer Nikhil Deshpande authored it. Its purpose is to help local governments experiment while protecting residents’ information and keeping government accountable for decisions involving the technology.

A service problem comes before a software purchase

The resource grew out of requests from local officials. At a Google-hosted roundtable in Atlanta earlier in 2026, state Rep. Dar’Shun Kendrick heard communities asking for help navigating AI. She asked the authority to adapt its state-agency guidance for local governments, Government Technology reported.

The playbook’s eight-step project process begins with defining a problem and considering alternatives—not selecting a vendor. A slow service might need a better form, an updated website or a change to an existing process rather than AI. Risk assessment, selecting or building a solution, testing and deciding whether to expand come later.

Officials are also urged to inventory what employees already use, including tools staff adopted on their own. That search extends beyond chatbots to AI embedded in permitting, document management, human-resources and cybersecurity software. Vendor reviews should examine hosting, the models involved and how information moves through the systems.

AI should support public employees, not replace the judgment, accountability, and community knowledge that effective government requires.

Shawnzia Thomas, Georgia state CIO, writing in the playbook

Review rises with the consequences

The guide separates projects into four categories: low, moderate, high and prohibited. Drafting or summarizing approved information with employee review can receive simpler controls. Moderate-risk projects go to a cross-functional AI committee; high-risk projects require a more extensive assessment and executive approval. Prohibited uses should not proceed unless the governing body formally changes its policy after legal and public review.

Systems involving law enforcement, surveillance, public safety, sensitive information at scale, or decisions affecting rights and access to services should automatically receive high-risk review. AI should not be the sole decision-maker on benefits, employment, enforcement, safety or access to government services. Qualified employees should check AI-generated information.

AI agents—systems that can take actions rather than only generate information—get additional controls. The playbook recommends starting with limited permissions, requiring human approval for consequential actions, recording detailed activity logs and maintaining a way to quickly disable the system.

Three phases, not a countdown

A proposed 90-day starting plan translates the recommendations into three phases. Deshpande stressed that the 30-, 60- and 90-day marks are not strict deadlines: communities can move at a pace that matches their resources.

1. Assign responsibility. Name an AI leader and coordinator, form a review committee, issue interim employee guidance and begin the inventory of existing tools.
2. Make review repeatable. Establish risk levels, vendor requirements and basic employee training, alongside processes for evaluating proposed uses.
3. Begin limited pilots. Launch a small number of projects, publish an initial AI webpage and establish incident response and reporting.

Local governments can adapt the included use-case intake form, vendor questions, policy examples and training recommendations. The guide also encourages a public register of significant AI uses so residents can see where the technology is involved in services.

The playbook is a resource, not a mandate to adopt AI. Kendrick wants local officials to weigh the benefits and risks for their residents, without feeling pressured to move faster than they are ready.