Classified cloud capacity is coming online for national security agencies just as agentic AI moves from answering questions to planning and executing multi-step tasks on its own. Because AI can now take actions instead of only surfacing insights, readiness and governance have become urgent.

In this second installment of a three-part series, Samantha Fletcher, Head of National Security Programs at AWS, continues the conversation with Dr. Adita Karkera, Chief Data Officer (CDO), Government and Public at Deloitte.

Fletcher starts by getting the data house in order through unified platforms, governance that enables AI adoption, workforce training and early prototyping in unclassified environments. Pointing to the recent launch of Amazon Bedrock AgentCore in GovCloud, Fletcher sets out three non-negotiables for agentic AI. Those are bounded autonomy, auditability and human authority over consequential decisions.

The conversation turns personal on ISR. Fletcher started as an unmanned aerial systems intelligence analyst and later built counter-UAS systems in the private sector. Processing drone, satellite and sensor feeds at the edge now lets AI flag critical intelligence within seconds instead of hours.

• Readiness before capacity: Agencies that unify data and prototype early can scale as classified capacity arrives.
• Autonomy with boundaries: Mission owners set explicit limits on AI agents, and those limits are enforced automatically.
• Humans keep consequential calls: AI can accelerate most of the workflow, while irreversible decisions stay with human judgment.
• Analysts move to higher-order work: Edge speed, cloud scale and AI sense-making shift analysts toward validating machine-generated insights.

Did you miss the previous episode?

On the move? Take this conversation with you. If you don’t have time to finish the video, you can stream the complete, uninterrupted audio edition of this episode on the go.