{"id":773155,"date":"2026-05-11T10:50:39","date_gmt":"2026-05-11T17:50:39","guid":{"rendered":"https:\/\/www.esri.com\/about\/newsroom\/?post_type=arcuser&p=773155"},"modified":"2026-05-11T10:50:39","modified_gmt":"2026-05-11T17:50:39","slug":"gis-governance-for-the-ai-era","status":"publish","type":"arcuser","link":"https:\/\/www.esri.com\/about\/newsroom\/arcuser\/gis-governance-for-the-ai-era","title":{"rendered":"GIS Governance for the AI Era"},"author":6921,"featured_media":0,"menu_order":0,"template":"","format":"standard","meta":{"_acf_changed":false,"sync_status":"","episode_type":"","audio_file":"","transcript_file":"","podmotor_file_id":"","podmotor_episode_id":"","castos_file_data":"","cover_image":"","cover_image_id":"","duration":"","filesize":"","filesize_raw":"","date_recorded":"","explicit":"","block":"","itunes_episode_number":"","itunes_title":"","itunes_season_number":"","itunes_episode_type":"","_links_to":"","_links_to_target":""},"categories":[25012],"tags":[493487,1741,241,490392,493486],"arcuser_issues":[493467],"class_list":["post-773155","arcuser","type-arcuser","status-publish","format-standard","hentry","category-managers-corner","tag-agentic-ai","tag-ai","tag-gis","tag-governance","tag-large-language-models","arcuser_issues-spring-2026"],"acf":{"short_description":"Advances in AI have created new and unfamiliar governance challenges. How can your organization best equip itself to handle these changes? ","pdf":{"host_remotely":false,"file":"","file_url":""},"flexible_content":[{"acf_fc_layout":"content","content":"I\u00a0wrote\u00a0an article for the fall 2022 issue of ArcUser<\/i>\u00a0called \u201cGIS Governance Distilled,\u201d which outlined a simple GIS governance framework and offered guidance on using it to improve the oversight of GIS programs. It was built on the notion that the most effective GIS departments are diligent about building a system of rules, practices, and processes to direct their efforts.\r\n\r\nIn the few short years since that article appeared, generative AI and other AI-based technologies have upended the GIS world, introducing solutions like smart assistants and autonomous agents. These advances have created an entirely new class of governance challenges driven by the trend toward machine-led decision-making and concerns about trust and reliability.\r\n\r\nThe simplified framework from the original \u201cGIS Governance Distilled\u201d article (outlining structure, controls, processes, and performance across six domains)\u00a0remains\u00a0a solid foundation.\u00a0What\u2019s\u00a0changed is the content within those components.\r\n\r\nWe need new roles or sets of responsibilities in the structure for AI oversight, new or revised controls (policies and standards) to guide AI usage and ethics, updated processes to handle the AI life cycle (from approval to audit), and new performance metrics (like measuring AI model accuracy or bias as part of program success).\r\n\r\nSo, how exactly is AI driving new governance best practices, and how can you address the shift?\r\n

Structure: New AI Oversight Functions<\/h2>\r\nAI introduces culpability and decision-making dynamics into GIS programs that traditional governance structures\u00a0weren\u2019t\u00a0designed to handle.\r\n\r\nFor example, when an AI-supported system provides a spatial analysis or generates a map autonomously, who is accountable if something goes wrong? Existing governance typically assigns a person (e.g., a data owner or application owner) to\u00a0be responsible for\u00a0those outcomes, but with AI, algorithms are increasingly acting partly on their own. This can lead to finger-pointing unless roles are updated. New oversight roles are\u00a0emerging\u00a0as a solution.\r\n\r\nGeneral IT guidance now calls for\u00a0establishing\u00a0an AI review board. This is a multidisciplinary committee that reviews and guides AI use, including legal, ethical, data science, and business stakeholders. Its duties would include oversight of bias,\u00a0approval of high-impact AI deployments, and monitoring for ethical compliance. This certainly applies to GIS solution deployment and would ideally include specialist GIS\u00a0expertise. Alternatively, consider embedding AI\u00a0expertise\u00a0into existing GIS governance committees. This ensures that whenever AI is part of a decision, someone at the table understands it and owns the consequences.\r\n\r\nUpdate committee charters to include AI responsibilities. For example, the GIS data subcommittee should oversee the quality and bias of AI training data, while the technology subcommittee should review AI system architecture and security.\r\n\r\nAdditionally, the role of a model owner or steward is\u00a0emerging\u00a0generally. IBM\u2019s guidance on data governance for generative AI suggests\u00a0designating\u00a0an owner for each relevant AI model who\u00a0is responsible for\u00a0the model\u2019s generation and operation. They would ensure that the model is developed and deployed in line with governance policies. For spatial machine learning models and spatial vision-language models, a model owner experienced with training spatial models should be assigned ownership and responsibility for the integrity and alignment with overall AI standards.\r\n

Controls: AI Embedded into Spatial Data Policies<\/h2>\r\nConventional GIS governance controls such as policy documents and usage guidelines often lack enforceable mechanisms for specific AI\u00a0behavior\u00a0and may need to be revised in the AI era.\r\n\r\nMany organizations have rules on how internal data can be shared or published, but they may not address scenarios like an employee feeding proprietary geospatial data to a generative AI service to obtain an analysis. Such an action could inadvertently expose sensitive data to an external model. Without clear policies, staff might not realize this is prohibited.\r\n\r\nOne tactic would be to implement a responsible AI-use policy or to update existing data policies to explicitly cover AI and GIS data specifically. This could include rules about what types of data can or can\u2019t be used to train AI or be inputted into third-party AI tools; requirements that only approved, secure AI platforms be used for certain data; and guidelines on reviewing AI-generated content for sensitivity before release.\u00a0It\u2019s\u00a0also\u00a0a good idea\u00a0to embed fairness and bias-mitigation standards into model development and deployment workflows.\r\n\r\nSimilarly, security policies should cover GIS deployments that access external large language model (LLM) services or require agents to have elevated system access. Ensure that AI systems undergo security testing and that any AI with elevated system access has constraints (an agent\u00a0shouldn\u2019t\u00a0be granted admin rights beyond its needs, for example). Agentic AI can chain actions in unpredictable ways, so enforcing the principle of least privilege (giving AI only the minimum access needed) becomes a procedural must.\r\n

Processes: Dynamic Oversight Needed<\/h2>\r\nGovernance\u00a0isn\u2019t\u00a0just about who\u00a0decides,\u00a0it\u2019s\u00a0also about how decisions are made and enforced day to day. AI introduces several new governance process requirements and puts pressure on existing ones.\r\n\r\nMost GIS departments have processes for evaluating new projects or technologies\u2014for instance, an architecture review or a project approval checklist. Historically, these might check for budget, alignment with strategy, security, etc. Now,\u00a0they must also ask: Have we evaluated the AI-related risks? AI systems can carry unique risks like algorithmic bias, unpredictable\u00a0behavior\u00a0such as hallucinations, or regulatory compliance issues (e.g., does using a cloud AI service violate privacy laws or data sovereignty policy?). If governance processes\u00a0don\u2019t\u00a0explicitly include these considerations, unintended risks could slip through. Project intake workflows should include AI risk screening, which asks whether a proposed solution uses AI, what kind of AI it uses, and whether it requires ethical or compliance review.\r\n\r\nMonitoring for data sovereignty adherence is particularly important. AI models trained on sovereign GIS datasets may embed sensitive spatial patterns or infrastructure details into their parameters. If these models are shared or commercialized, it could result in indirect exposure of protected data even if the raw data\u00a0isn\u2019t\u00a0explicitly shared. This is especially problematic for sensitive land data, critical infrastructure, and\u00a0defense-related geospatial assets.\u00a0It\u2019s\u00a0worth implementing private-by-default AI strategies. Use local or on-premises AI models for sensitive GIS data, or leverage\u00a0retrieval-augmented generation (RAG) architectures that keep data within sovereign boundaries.\r\n\r\nAlso, generative AI\u00a0is capable of producing\u00a0new spatial content (e.g., synthetic maps, inferred land-use layers) based on sovereign data. If these outputs are used outside the originating\u00a0jurisdiction\u00a0or without proper governance, they may undermine local control over how spatial knowledge is represented and used. Define rules for how synthetic or inferred geospatial outputs can be used and shared, especially when derived from sovereign datasets.\r\n\r\nChange management must account for AI model updates. If an AI model is retrained, governance should define who approves the\u00a0new version\u00a0and how\u00a0it\u2019s\u00a0tested.\r\n

Performance: AI-Specific Metrics<\/h2>\r\nAs AI becomes increasingly embedded in GIS programs, traditional performance governance must evolve to address new dimensions of success, accountability, and risk.\r\n\r\nHistorically, GIS program performance was measured with metrics like system uptime, data quality, and user satisfaction. However, AI introduces complex, dynamic\u00a0behaviors\u00a0such as autonomous decision-making, generative outputs, and model drift that require more nuanced and continuous oversight.\r\n\r\nOne major challenge is redefining what success looks like. AI-driven GIS outputs must be evaluated not only for technical accuracy but also for fairness, transparency, and ethical compliance. For example, a model with high overall accuracy may still produce biased results that disproportionately affect certain communities. Governance must incorporate fairness audits, bias detection, and stakeholder trust metrics to ensure\u00a0equitable\u00a0outcomes. Augment your existing GIS performance indicators with AI-focused ones. Consider metrics for accuracy, bias, user trust, and AI uptime. Implement continuous monitoring.\u00a0Don\u2019t\u00a0treat AI evaluation as a one-off. Many AI platforms can\u00a0monitor\u00a0whether input data is shifting or whether outputs start differing from expected patterns. Subscribe to these alerts.\r\n\r\nAI-based decisions must also be explainable and traceable, especially when used in planning, public services, or regulatory contexts. Governance frameworks should track how often AI outputs are overridden by human reviewers, whether explanations are available, and how well ethical principles are embedded into workflows. Integrate AI performance into regular governance reporting. For example, in quarterly reports to\u00a0your GIS steering committee, include a section like this: \u201cAI in our operations\u2014here\u2019s what it did, here\u2019s how it performed, here are any issues and how we addressed them.\u201d This keeps AI\u2019s contributions and challenges visible at the executive level, which is important for accountability.\r\n\r\nOn the technical side, performance monitoring also becomes more complex. AI models can degrade over time due to data drift or changing conditions. GIS governance must implement continuous validation and model-retraining protocols. Dashboards should report on model accuracy and compliance status,\u00a0similar to\u00a0how system health is tracked today. Some of this means\u00a0it\u2019s\u00a0important to invest in monitoring infrastructure, such as audit logs and centralized dashboards, which provide visibility into AI\u00a0behavior.\r\n\r\nIn addition to numbers, gather qualitative feedback. Have an open channel for users to report concerns or odd\u00a0behaviors\u00a0from AI tools. Track how many suggestions for improvement come in related to AI. This can be a performance indicator of its own (if too many people report that the AI\u00a0isn\u2019t\u00a0helpful,\u00a0that\u2019s\u00a0a problem to solve).\r\n

Integrating AI into Your GIS Strategy<\/h2>\r\nI\u2019d\u00a0be remiss if I\u00a0didn\u2019t\u00a0mention strategy. Many GIS strategic plans did not originally account for\u00a0AI, or\u00a0treated it only as a distant innovation topic. Now that generative and agentic AI are becoming mainstream (with GIS software incorporating AI capabilities and users expecting them), not having a clear AI direction is a governance gap.\r\n\r\nOrganizations may end up with ad hoc AI experiments (some departments forging ahead, others holding back) that\u00a0don\u2019t\u00a0align with long-term\u00a0objectives. Governance at the strategic level should ensure that there is a cohesive AI game plan: either as part of the GIS strategy or as a stand-alone AI strategy that interlocks with it.\r\n\r\nYour strategy should answer these questions:\r\n