Security at machine speed has its growing pains

CTO of the Department of the Navy Justin Fanelli, Vice Admiral (Ret.) Timothy James White, Chief of Texas Cyber Command, and Jamie Wolff, CIO of the National Nuclear Administration, sometimes sparred over AI security topics and risk assessment, but agreed that cybersecurity isn't moving at machine scale yet.

The panel at the Google Public Sector Summit 2026 followed a move to launch of Wiz for Government. Wiz, which is part of Google Cloud's security portfolio, now has Wiz agents designed to project AI models and applications in FedRAMP High environments. The effort is focused on post-quantum cryptography. Wiz Red, Green and Blue agents are now available for Wiz for Gov.

• Google Public Sector customers: Data foundation eases AI bottlenecks
• Google Public Sector launches Gemini for Government Services, sees AI agents as integration accelerator
• Accenture Federal Services, Google Public Sector roll out Gemini Launchpads

Here's a look at the takeaways from the panel.

Human speed security isn't fast enough to secure AI agents. Fanelli said that the agentic AI era is a speed context that will force the public sector to work more closely with the private sector. "Human speed, even fast human speed, is insufficient and that requires the best public-private partnership we've ever seen," he said.

Wolff said adoption of AI security is more of an issue than the threats that are emerging. "I'm actually more concerned about our ability to adopt change at the speed necessary for superintelligence," said Wolff.

Do perimeters matter? Fanelli said perimeters don't matter as much in AI security. Wolff disagreed. "I don't think traditional security perimeter security has gone away, but now we have these additional things that we have to secure with the data, the endpoints and the deployed systems," he said.

White said boundaries are often set by law and his approach starts with observability. White oversees the first state to have its own cybersecurity command. "Fundamentally you have to be able to see your terrain so you have the capability to share telemetry from all of your constituencies," said White.

Agents need to be secured like all assets. Ron Bushar, Chief Information Security Officer and Managing Director of Google Public Sector, said security management has evolved from physical assets to networks to data and now AI agents. "You are going to have agents that are acting on behalf of you and your policies and authorizations, and you've got to manage where they go, what they can do and audit what they're doing," said Bushar.

Risk management. Fanelli argued that how organizations think about risk has to change. "The riskiest position is staying still and most people still define risk locally," he said. "This is a people problem."

Fanelli said organizations need to go for the win and not just avoid mistakes.

Wolff pushed back. "Our mission is to produce nuclear weapons so I can't insert additional technology risk by poor decision-making," he said.

White took the middle ground and said risk tolerance depends on the function. Also keep in mind that adversaries are going for wins.

Critical infrastructure is vulnerable. White said there are small utilities with 30-year-old equipment and little budget. Protecting those assets will take national and state frameworks to defend. "No water, no life. No power, no telecom," he said. "You have to bind these assets together.

The risk is even higher given that humans can't keep up with patches. Detection, prioritization and patching will have to be automated and carried out by AI agents. "In the past, you had the comfort of time. That's gone. Literally, I know someone is coming after me tomorrow," said Bushar.