ICE work with AI agents is minimal, CIO says

Immigration and Customs Enforcement has led the adoption of artificial intelligence for the Department of Homeland Security, but the unit is taking a slower pace with AI agents, according to the component’s top IT official. 

“We are not using a lot of those features,” ICE CIO Dustin Goetz said during AFCEA Bethesda’s LEAPS Summit last week in Washington, D.C. “And we’re not planning on it.”

AI agents are the subject of immense hype — and confusion. Even as technology vendors trumpet the technology’s potential, there is no universally agreed-upon definition of an AI agent, illustrating the technology’s immaturity. The muddy marketing tied to the high-risk, semi-autonomous capabilities can spell trouble for federal agencies. 

“The ability to predict, isolate, and manage the risks of agentic systems that can reach out into the world is not a solved problem,” said Todd Jacobs, co-founder of the D.C.-based think tank Theia Institute, which focuses on emerging technologies, cybersecurity, AI ethics and governance. 

Probabilistic agents, he added, “take on tasks defined for them.”

“But ‘how’ they complete the tasks — and in some cases, whether they even follow the instructions provided — can’t be guaranteed even by the AI companies that build the agents and models,” Jacobs said in an email to FedScoop. 

Standard-setting organizations are working to fill in the gaps. The National Institute of Standards and Technology launched an AI Agent Standards Initiative in February to begin cultivating best practices and protocols for implementation, but there isn’t yet an agentic-specific framework for agencies to follow. 

Still, some agencies are beginning to get their feet wet. 

The State Department is in the “exploration stage” of agentic AI with an eye on reducing administrative burden, according to acting Chief Data and AI Officer Amy Ritualo. In sandboxed environments, State is testing the tool’s ability to sort through malware. 

The Department of Health and Human Services listed agentic AI use cases in its 2025 inventory, and the Centers for Disease Control and Prevention incorporated the technology in its AI strategy documents earlier this year. NASA is an early adopter, too, using the tech to prepare risk-based recommendations of whether a mission is ready to fly safely. 

The FBI, on the other hand, is taking an approach more aligned with the hesitation voiced by IT leaders at ICE. 

“Our folks, when they hear ‘agentic workflow,’ we all get super excited, and then we come back to reality,“ Katie Noyes, chief AI officer at the FBI, said during the event last week. “The tools that are commercially offered, and even some that are oriented to those of us in government, we need a little bit more transparency, and that’s where some things have been challenging.”

For agencies moving forward on agentic AI, the preparation checklist for adoption is long. 

“True readiness requires more than technological enthusiasm,” said Dia Adams, board chair at The AI Table, a D.C.-based think tank. “It demands robust oversight frameworks that define clear intervention triggers, accountability mechanisms, and guardrails.”

“Agencies must also ensure their data architectures are AI-ready: integrated, clean, and accessible in real time, so that agentic systems can operate across workflows without creating silos or security vulnerabilities,” she added. “It’s important to ensure that deployment doesn’t outpace the development of the foundations needed to support these systems at scale.”

What AI is doing for ICE now

AI agents might be taking a backseat at ICE, but artificial intelligence more broadly is driving results. 

ICE is working on nearly 45 AI use cases, representing nearly one-quarter of DHS’s total count, according to its latest inventory. More than half of those use cases cropped up in the six-month period following July 2025. 

The DHS unit credited its technology efforts for helping in-the-field agents more than double a fiscal 2025 arrest goal.

Palantir is one of the more controversial technology vendors helping ICE execute its mission. The company sells software that helps agencies process and analyze data, an increasingly important part of day-to-day operations for DHS and its growing AI stack. 

“As far as AI and where we are right now, Palantir is our big partner,” ICE Assistant Director Matthew Elliston said during the event last week. “Palantir and AI is not telling us who to arrest, but they are giving us leads and saying, ‘hey, here’s all the information about this person.’”

Palantir is powering ICE’s current “targeting system” that has an 80% success rate, per Elliston, meaning that 4 out of 5 times agents “walk out with somebody in handcuffs” when the system identifies a location of interest. The previous homegrown system, he said, only had about a 30% success rate. 

“User adoption was really difficult because we had just lost the trust of all of our 20,000 people in the field by giving them a terrible product first,” Elliston said. 

While the partnership between ICE and Palantir is critical to the current operations, it’s also supporting the agency’s ability to remain agile enough to defend against emerging threats of the future, according to the official. 

“As they address new threats and changes and as their system grows, that’s all contractually part of ours,” Elliston said.