The American Immigration Council is a non-profit, non-partisan organization. Earlier this year, the Department of Homeland Security (DHS) released an updated AI Use Case Inventory. Required by federal law and Office of Management and Budget guidance, the inventory reveals where DHS is investing, how quickly new systems are being deployed, and the growing role AI is playing in immigration enforcement.
Compared to last year’s inventory, the data shows that DHS is rapidly expanding its use of AI in immigration-related operations. Between December 2024 and January 2026, active immigration-related AI use cases increased from by 36%. During just the six-month period between July 2025 and January 2026, DHS reported 58 new AI use cases. While some older systems were retired or consolidated, the overall trend is clear. DHS continues to expand its AI footprint.
Most of that growth comes from immigration enforcement agencies. Immigration and Customs Enforcement (ICE) added 28 new AI systems, while Customs and Border Protection (CBP) added 23. Together, these two agencies account for nearly 90 percent of the new immigration-related AI programs. In contrast, relatively few new systems focus on benefits processing or customer service. Most support investigations, surveillance, targeting, or other law enforcement activities.
These systems are also reaching deployment quickly. Of the 58 new use cases, 27 were already deployed and three were in pilot programs. Only a minority remain in earlier stages of development, suggesting DHS is moving AI systems into operational use at a rapid pace.
The inventory also shows that generative AI has moved beyond experimentation. Half of the newly reported systems, 29 out of 58, use generative AI. Across DHS, these tools are being used for document summaries, semantic search, investigative support, translation, code generation, and other operational tasks. Just a year ago, much of the public discussion focused on predictive analytics and computer vision. The latest inventory suggests generative AI is quickly becoming part of everyday government operations.
This shift is significant because generative AI creates new content rather than simply analyzing existing data. It generates new output, including summaries and analyses that may shape how personnel interpret a case. While these systems are designed to assist rather than replace human decision-makers, research has shown that people do not always detect AI-generated errors or hallucinations. As generative AI becomes more integrated into immigration enforcement, effective human review depends not only on having a person in the loop, but also on ensuring reviewers have the time, training, and access to the underlying information needed to identify mistakes.
ICE’s use of AI has already led to the wrongful detention of both immigrants and protestors. Mobile Fortify, a facial recognition app utilized by ICE, has misidentified individuals for enforcement operations.
The updated inventory also highlights an important shift in how DHS classifies AI systems. Under Office of Management and Budget (OMB) guidance, high-impact AI includes systems that play a major role in decisions affecting a person’s rights, access to government services, or safety. Because of the potential consequences, these systems are subject to additional oversight and testing.
In the latest inventory, DHS designated 15 new systems as high-impact. Another 14 were classified as “presumed high-impact but determined not high-impact.” In other words, these systems initially appeared to meet the definition of high-impact AI, but after review, DHS concluded they did not. In nearly every case, DHS explained that the AI supports, but is not the principal basis for, an enforcement or operational decision because a human remains involved.
That distinction matters because it determines which federal oversight requirements apply. As DHS relies more on generative AI to summarize, search, and analyze information, the effectiveness of human review becomes increasingly important.
The types of systems receiving this designation help illustrate why the “presumed high-impact but determined not high-impact” distinction matters.
ICE’s Enhanced Lead Identification and Targeting, developed with Palantir, uses generative AI to organize investigative leads and support enforcement operations. Another ICE program, Open-Source Intelligence for Lead Identification and Targeting, analyzes publicly available online information to identify potential investigative leads. Meanwhile, Semantic Search and Summarization for Investigative Data, helps investigators search years of reports and automatically summarize findings. CBP’s RAPTOR system combines computer vision, radar, infrared imagery, and other sensor data to detect and classify activity along the border.
Although these systems support investigations, surveillance, and operational decision-making, DHS concluded they are not high-impact because a human remains responsible for the final decision.
The DHS AI inventory is an important transparency tool, but it is only a starting point. Many systems include only brief or vague public descriptions, vendor information is often incomplete, and information about testing, bias assessments, accuracy, and outcomes remains inconsistent. As AI becomes increasingly embedded in immigration enforcement, meaningful transparency will require more than simply listing systems. The public also needs a better understanding of how these tools are evaluated, how they affect decision-making, and what safeguards exist when they make mistakes.
