DHS-2515 – AI-Enhanced ICE Tip Processing

DHS-2515

AI-Enhanced ICE Tip Processing

ICE

Deployed Not high-impact Generative AI Law Enforcement

Problem Statement

This use case intends to solve the problem of the time-consuming manual effort required to review and categorize incoming tips.

Expected Benefits

The use of AI in this process enables the Tip Line team to more quickly identify and action tips recommended for urgent case categories. Additionally, the introduction of a BLUF field saves time by providing analysts with a high-level understanding of a tip before they review its details.

System Outputs

This solution uses a large language model (LLM) to enrich web tips with two additional data elements: (1) a high-level summary of the tip (BLUF), and (2) a recommended case category. The LLM generates BLUFs in English, regardless of the language used in the raw tip submission. For non-English tips, analysts may click a button to translate the full tip violation summary data element into English. The LLM is configured to only recommend case categories from a list of predefined HSI case categories.

Documentation

Operational Date: 5/2/25

Procurement: a) Purchased from a vendor

Vendor(s): Palantir

ATO: Yes

System Name: Case Management & Analytics (CMA)

Data & Code

Training Data: The system uses commercially available large language models trained on the public domain data by their providers. There was no additional training using agency data on top of what is available in the models’ base set of capabilities. During operation, the AI models interact with tip submissions.

PII Involved: Yes

Demographic Variables: None

Custom Code: Yes

US Department of Homeland Security