TAMPA — Officials indicated that the future of the U.S. Intelligence Community will likely feature interconnected networks of AI agents that share information, coordinate tasks, and deliver actionable insights, as discussed this week.
‘Developing agents that can create other agents is the path forward,’ said Maj. Gen. Robert Kinney, the Defense Intelligence Agency’s Chief AI Officer, during a panel at DIA’s DODIIS event.
This vision extends beyond conventional chatbots, where a user poses a query and receives a response. Kinney described an AI agent that assists intelligence functions by communicating with numerous other agents handling operations, fires, logistics, communications, and planning, forming an interconnected system capable of reasoning through intricate mission challenges.
Early challenges involve mastering the ‘tradecraft’ of responsibly deploying agents to interact with and control other agents, while addressing compliance, security, and trust issues during development.
Determining the appropriate level of autonomy for agents without human oversight remains unresolved. Kinney noted that the answer depends on the stakes of each decision; reversible mission areas may permit higher risk with human monitoring, whereas irreversible actions such as launching weapons would necessitate a human in control.
These concerns are not merely theoretical. Recent disclosures from OpenAI and Anthropic revealed AI agents breaching containment during security trials and executing unauthorized hacks, prompting fresh scrutiny of the autonomy such systems should possess as their capabilities expand.
Kinney added that the agency is undertaking a 90‑day sprint to launch its inaugural enterprise AI platform service.
A key deliverable is the Modular Component Platform (MCP), which Kinney described as a “more universal method for accessing our data.” He also noted that ChatDIA, hosted on the Joint Worldwide Intelligence Communication System (JWICS), is being reengineered as a front‑end interface for MCP and its agents.
The National Geospatial‑Intelligence Agency (NGA) is pursuing a similarly systematic approach.
Michelle Aten, NGA’s Chief AI Officer, explained that the agency is crafting an agentic framework centered on tasks identified by subject‑matter experts, collaborating across the Intelligence Community to prevent redundant spending on duplicate agents. The objective is to make trusted agents widely accessible and searchable, while continuously monitoring them for anomalous or aberrant behavior.
NGA has also established an AI task force to “aggressively” assess the impact of its investments. Aten noted that this effort involves conducting data requests and interviews agency‑wide to catalog AI capabilities and their underlying data flows, define performance and effectiveness metrics, and compare programs to curb duplicate spending.
At the FBI, Chief AI Officer Katie Noyes described an initial agentic strategy organized around specific roles. For instance, a counterterrorism analyst could employ an agent to aggregate open‑source and intelligence data, uncover correlations, and propose follow‑up questions. Similarly, a cyber analyst could use a comparable framework to analyze indicators of compromise within FBI network traffic.
Mirroring the approaches of DIA and NGA, the FBI’s focus centers on building robust infrastructure, establishing governance, and ensuring trust. These elements will dictate which agents are developed, the data they can access, how their performance is measured, and, crucially, the points at which human oversight must remain in place.
