Artificial intelligence (AI) could help governments speed service delivery, modernize aging software, and strengthen cyber defenses, but agencies need to distinguish routine tasks suited to experimentation from consequential decisions that demand greater scrutiny, government and industry leaders said Nov. 10.
Speaking at the National Academy of Public Administration’s 2026 National Conference in Washington, D.C., Tempe, Ariz., Mayor Corey Woods and leaders from OpenAI and Anthropic urged governments to pair AI adoption with clear safeguards and human oversight.
Tempe looks to AI to improve city services
Woods said Tempe adopted an ethical AI policy in 2023 after city employees had already begun using dozens of AI tools. The policy, developed largely by city staff, including CIO Jared Morris and his team, was intended to provide transparency about how the technology is used and what safeguards are in place.
“We needed to adopt an ethical AI policy, so they know we’re not just simply, you know, willy-nilly using any program,” Woods said.
Woods pointed to permitting as one area where AI could improve city services, particularly for small businesses that may lack the resources to hire specialists to navigate government processes.
“Whatever we can actually do to help those folks get through the process a lot quicker, we want to do that,” he said.
The mayor also acknowledged public concerns about AI, saying government leaders should explain the technology’s potential benefits while encouraging employees to experiment responsibly.
“Do not be afraid to fail,” Woods said.
Industry leaders emphasize governance and human oversight
Industry panelists identified opportunities for AI to reduce routine work, help residents navigate government services, and accelerate research. They also stressed that agencies must adapt their governance practices as the technology evolves.
Alexis Bonnell, partnership manager at OpenAI, said organizations should regularly revisit their AI policies to account for changing capabilities and risks.
“If you are not updating your governance policy … on a monthly basis,” she said, “organizations risk missing both new risks and new opportunities to serve the public.”
Bonnell also emphasized the importance of keeping “people in control” of AI systems. She said users can tailor AI tools to their needs through carefully selected source materials and custom instructions.
Those capabilities can help address concerns about unreliable or biased outputs, she suggested, but they do not eliminate the need for organizational safeguards.
Thiyagu Ramasamy, head of Americas public sector at Anthropic, identified legacy system modernization and cybersecurity as additional opportunities for government AI adoption.
“There’s so much legacy code in the system that could be modernized. The permitting system could be really old. It’s much easier to actually rewrite it to modern technologies much, much faster,” Ramasamy said. Ramasamy said AI models could also help agencies strengthen their network defenses.
“Cyber defense … is becoming extremely critical across the federal, state, local, everywhere that we should be prioritizing it,” he said.
However, he cautioned that successful adoption requires more than giving employees access to AI tools. He recommended distinguishing between low-consequence tasks, where experimentation may be appropriate, and high-consequence decisions that require greater scrutiny.
Ramasamy described AI as a “chief of staff” that can organize information and support decision-making without replacing human judgment.