US Department of Homeland Security

AI Use Case Inventory

2025 Inventory

Source data: DHS
Glossary
DHS-2578

Enhanced Lead Identification and Targeting

ICE
Deployed Presumed high-impact but determined not high-impact Generative AI Law Enforcement

Problem Statement

The AI is intended to solve the problem of unstructured, hard‑to‑read address information in records like rap sheets and warrants, which makes it difficult and time‑consuming for Enforcement and Removal Operations officers to extract accurate addresses and build usable enforcement leads.

Expected Benefits

The integration of AI enhances data extraction capabilities and decreases the time spent on manual data normalization tasks. This provides Enforcement and Removal Operations officers with higher-quality leads and enables them to make better-informed decisions.

System Outputs

The outputs of Enhanced Leads Identification & Targeting for Enforcement (ELITE) are enriched leads that include AI-extracted addresses. Enforcement and Removal Operations officers review these leads to determine which are actionable and then share actionable leads across offices and areas of responsibility to coordinate enforcement operations.

Impact Justification

While ELITE provides actionable data to ERO officers, its outputs are limited to normalized address data and do not serve as a principal basis for decisions or actions with legal, material, binding, or significant effects on individuals. ERO officers review and validate the AI-driven outputs before determining actions, ensuring human oversight and additional verification steps. Furthermore, enforcement decisions are based on the full investigative process, which includes human analysis and validation of the source of AI outputs. As such, the AI system's role is limited to data extraction and normalization, rather than serving as a primary basis for enforcement actions.

Documentation

Operational Date: 6/7/25
Procurement: a) Purchased from a vendor
Vendor(s): Palantir
ATO: Yes
System Name: Palantir Federal Cloud Service

Data & Code

Training Data: The system uses commercially available large language models trained on the public domain data by their providers. The use of LLMs is limited to address extraction from criminal records such as rap sheets and warrants. ICE data was not used during the design, development, or training phases of the AI models. During operation, the AI models interact with ICE production data from multiple sources, including data from ICE’s Enforcement Integrated Database (EID).
PII Involved: Yes
Demographic Variables: None
Custom Code: Yes