Why ‘move fast and break things’ doesn't work for government AI

Artificial intelligence is moving into government at remarkable speed. It can answer resident questions, help employees analyze data, automate routine work and free the people who serve their communities to focus on work requiring human judgment, expertise and connection. 

Done well, AI can help government deliver faster service, make better decisions and serve residents more effectively.

But when it comes to government AI, speed isn’t the goal. Trust is. 

The private sector can move quickly and occasionally get things wrong. Government doesn’t have that luxury. Government AI failures can have life-altering consequences. An AI system giving incorrect information on public health guidelines or city code can endanger citizens. A misinterpreted regulation from an AI chatbot creates legal liability. An AI system endorsing a political candidate breaches public trust. When decisions involve public dollars, public safety or public trust, “the AI made a mistake” is not an acceptable explanation. 

Government must build trust from the start in the technology, the partner, the process and the people responsible for making sure AI works as intended. Safeguards cannot be bolted on after deployment. They must be built into the foundation from the beginning. 

That foundation means responsible AI adoption: building the systems, processes and accountability needed to make AI work at scale. That means sustainable change management, clear guardrails, meaningful oversight and human judgment at every stage to strengthen the people and institutions responsible for serving the public.

Government’s higher bar is rooted in the unique relationship between public institutions and the people they serve. Residents cannot opt out of government the way they can walk away from a private service. Government agencies hold highly sensitive personal information like tax records and health data, and residents are often required to provide it. When trust is lost, there is no competing provider to choose. 

Tackling the Governance Challenge 

The real challenge for government leaders is translating that higher bar into practical AI adoption: putting the right structures, processes and people in place to ensure AI is introduced thoughtfully, governed effectively and delivers value without compromising public trust.

States and municipalities are already asking hard questions. According to the National Association of State Chief Information Officers, 88% of states have established AI responsible use policies and governance frameworks. Key non-negotiables include:

• Data sovereignty: Absolute control over data, isolated from third-party AI and commercial models, given the sensitive nature of government information. 
• Vendor neutrality: The ability to switch providers if they aren’t delivering. 
• Rigorous testing and transparency: Vendors should provide granular evidence that systems refuse incorrect information, have been tested for bias and can handle regulatory nuance.

State and local leaders understand governance is essential, though many are still determining what it should look like. The National Institute of Standards and Technology provides a framework for critical infrastructure with guidance on trustworthiness characteristics like explainability, graceful degradation and human-in-the-loop oversight. 

Don’t wait for perfect governance. Instead, use frameworks like NIST’s as a starting point, anchor to best practices while maintaining flexibility, introduce checkpoints where processes can be adjusted and evaluate efficacy at every step. 

Different Rules. Different Playbook 

If government operates under different rules, it needs a different playbook. The answer isn't to slow innovation, but to approach adoption with the same intentionality we expect from government decisions every day. That means the right foundation: clear governance and accountability, thoughtful change management and human oversight. 

Here are the steps government leaders can take to move forward with confidence.

Identify the problem. AI trailblazers are prioritizing opportunities with meaningful impact but minimal risk. For example, agentic AI handling data entry, answering repetitive questions, routing routine approvals and managing workflows. These uses don’t revolutionize criminal sentencing or reimagine permitting systems on day one, but they do free employees to do other meaningful work. Low-risk, high-impact opportunities give governments a safe testing ground for governance structures. 

Know your data. AI is only as good as the data behind it. Bad data at machine speed produces bad outcomes at scale. Before AI touches anything, build an inventory: what data you hold, how sensitive it is, who owns it, where it lives and how long you keep it. 

This does double duty. It shows which data AI should use and it keeps you in control of everything else, including the compliance obligations that only the public sector carries. Agentic AI acts with real autonomy. An agent moving sensitive resident information somewhere it doesn’t belong isn’t a technical glitch; it’s a breach of public trust. 

Governance is good. Establish and anchor to established frameworks after understanding your problems and data. Governance doesn’t need to start from scratch. Good governance starts with best practices and is built upon distinct needs. 

Trust is key. Find a trusted vendor with deep expertise. The right partner can help you understand where AI delivers value and where it shouldn’t be used. Ask for evidence: deep expertise in your domain, a demonstrated record in public sector environments and references from agencies like yours. The right partner protects your data and helps you avoid costly missteps.

Relentless pursuit. Successful implementations demand rigor throughout the lifecycle. Invest in change management, communicate transparently about AI capabilities and limitations and reimagine workflows rather than overlay AI on existing processes. AI is an opportunity to rethink how work actually happens. 

Upskill. Start with high-impact, low-risk areas and build confidence and institutional knowledge to create momentum for the future. Everyone needs to upskill as AI evolves – and be accountable to AI outputs. Establish learning venues, including communities of practice and training modules, so everyone learns as the technology evolves. 

Progress Not Perfection

The risk of getting AI wrong is real, but so is the opportunity to get it right. Government leaders don’t need every answer before beginning and shouldn’t let the pursuit of perfection block progress. But they do need clarity about what all that work is building towards. Every step is a way to earn trust, and trust makes the next step possible.

Leaders who prioritize trust, establish adaptable frameworks, start with the right problems, select the right partners and prepare their people for change will be best positioned to capture AI’s benefits responsibly. They can move forward with confidence, safely learn as they go and adapt as the technology evolves, while keeping the focus where it belongs: on using AI to help government work better for the people it serves.

Franklin Williams is chief AI officer at Tyler Technologies.