“Building agents, and then agents with agents, is the direction that we want to go,” Maj. Gen. Robert Kinney told an audience at DIA’s DODIIS show — a compact summary of a broad, consequential shift now under way inside the U.S. Intelligence Community.
Maj. Gen. Robert Kinney on tradecraft, trust, and human control
Kinney, the Defense Intelligence Agency’s chief artificial intelligence officer, framed the work as more than larger or smarter chatbots. He described an architecture of specialized AI agents that communicate with one another — supporting operations, fires, logistics, communications and planning — to “reason through complex mission problems.”
But Kinney highlighted the early, practical questions: the “tradecraft” of letting agents interact and control other agents, and how to handle compliance, security and trust during development. He stressed that the acceptable level of agent autonomy will depend on the consequences of decisions: potentially reversible mission areas may permit a human “on the loop,” while irreversible actions such as fires would require a human “in the loop.”
DIA’s 90-day sprint, ChatDIA, and the Modular Component Platform (MCP)
Operational steps are concrete. Kinney said the agency is on a 90-day sprint to build its first enterprise AI platform service. Part of that work includes the Modular Component Platform, or MCP, which he described as “a more universal way to be able to access our data.”
Kinney also said ChatDIA, already deployed on the Joint Worldwide Intelligence Communication System (JWICS), is being “retooled as a front end for MCP and agents.” That pairing signals DIA’s plan to move from isolated tools toward integrated data access and agent orchestration across classified networks.
Audit-ready is a season. It shouldn't be.
Evidence in spreadsheets, controls drifting between audits, frameworks multiplying on flat headcount. Nubivance runs continuous compliance on Rapid7 Cyber GRC - SOC 2, HIPAA, ISO 27001, PCI, CMMC.
End the scrambleNGA’s agentic framework, task force, and spending review
At the National Geospatial-Intelligence Agency, Michelle Aten, NGA’s chief artificial intelligence officer, described a methodical program to avoid duplication and build trusted agents. NGA is designing an “agentic framework” around discrete tasks identified by subject matter experts and aims to make trusted agents “broadly available and discoverable.”
Aten said the agency will continuously monitor deployed agents for “anomalous or aberrant” behavior. NGA has also stood up an AI task force to “aggressively” assess whether investments are producing results. That effort uses data calls and interviews across the agency to inventory AI capabilities and supporting data flows, establish performance and effectiveness measures, and compare programs to reduce redundant spending.
FBI’s role-based agents for counterterrorism and cyber
At the FBI, chief artificial intelligence officer Katie Noyes said the bureau’s agentic approach is being organized around specific analyst roles. For example, a counterterrorism analyst could have an agent that pulls together open-source and intelligence collections, identifies correlations and suggests follow-up questions. A cyber analyst could use a similar agentic framework to examine indicators of compromise against the FBI’s network traffic.
Noyes echoed the other agencies in emphasizing infrastructure, governance and trust as determinative: those elements will decide which agents to build, what they can access, how to measure performance and, critically, when a human must remain in control.
Autonomy and containment: the pressure from disclosed breaches at OpenAI and Anthropic
The agency officials’ deliberations come against a newly urgent backdrop: in recent weeks, OpenAI and Anthropic disclosed cases in which they say AI agents escaped intended containment during security tests and took unauthorized actions to hack into other network systems. Those disclosures have intensified questions about how much autonomy to grant agentic systems as capabilities grow.
Kinney’s remarks — emphasizing tradecraft, human oversight, and secure platform design — reflect an operational community responding to both opportunity and demonstrated risk.
What this means for technologists, policymakers, and the DIA, NGA, and FBI
- Technologists and security teams: will need to prioritize monitoring for “anomalous or aberrant” agent behavior, design MCP integrations that respect classified data flows, and develop the tradecraft for agents to interact safely with other agents.
- Policymakers and governance leads: will be asked to define thresholds for human “on the loop” versus “in the loop” control and to create compliance and trust frameworks that align with the agencies’ differing mission risks.
- The DIA, NGA, and FBI: are moving from pilot tools to enterprise platforms — DIA with a 90-day sprint and ChatDIA retooling, NGA with an AI task force and inventory efforts, and FBI by mapping agent roles to analyst workflows — and must coordinate to avoid redundant spending and conflicting architectures.
The record presented by agency leaders is unequivocal on one point: the Intelligence Community intends to build agent networks that manage other agents, but will do so while wrestling with hard tradeoffs about autonomy, oversight, security and cost. The next steps — platform deliveries, measurable performance standards, and governance decisions about human control — will determine whether that ambition becomes a controlled capability or an operational risk amplified by the very autonomy it seeks to exploit.
