Skip to content

Meet the DHS chatbots

Components across the Department of Homeland Security now run their own AI assistants. Open any one and it will tell you what its public filing says.

Chatbot field guide

8chatbots on the list
5deployed or in pilot
2answer Yes on personal data
0rated high-impact by DHS

Counts come from the 8 assistants listed below, as DHS reported them in its June 12, 2026 filing. The high-impact rating is DHS's own determination.

8 of 8 chatbots shown

ICE's STELLA is more than a chatbot

ICE's procurement notice describes five layers. The inventory lists only the assistant, as pre-deployment. Open each layer.

STELLA platformDescribed as running

A multi-cloud platform on Google, Microsoft Azure and AWS.

“secure access to frontier models across cloud environments”

ICE RFI-ICE_ADTS_Sep2026, section 2.0

STELLA ChatDescribed as running

An agency-wide chatbot with web grounding and retrieval from knowledge bases.

“an enterprise-wide, ICE-branded GenAI chatbot”

ICE RFI-ICE_ADTS_Sep2026, section 2.0

Application portfolioDescribed as running

ICE says twelve applications are deployed.

“currently actively supporting ICE's operational and investigative missions”

ICE RFI-ICE_ADTS_Sep2026, section 2.0

Agentic Software FactoryFuture state

A planned layer to coordinate AI agents.

ICE describes this layer as a future-state concept, not an existing implementation.

ICE RFI-ICE_ADTS_Sep2026 (summarized, not quoted)

Authority to OperateTwo answers

The RFI says STELLA has a full ATO. ICE's Oct. 21 answers to vendor questions say interim.

“STELLA and associated applications have an interim ATO.”

ICE answers to vendor questions, posted Oct. 21, 2026

The rules these tools run under

Oversight trigger“DHS requires use of AI to have human oversight when the use is high impact”DHS Directive 139-08 Rev. 01, effective Sept. 3, 2026
Support-function carve-outThe directive says relying on AI for a support function such as language translation is not relying on AI outputs as a sole basis for agency action.DHS Directive 139-08 Rev. 01 (summarized, not quoted)
WaiverThe directive lets the Chief AI Officer waive any requirement, including prohibited uses, consistent with law.DHS Directive 139-08 Rev. 01 (summarized, not quoted)
OMB presumption“Provision of language translation ... when responses are legally binding or for an interaction that directly informs an agency decision or action”OMB M-25-21, Appendix, section 6(o)

Sources and method

Each assistant comes from DHS's June 12, 2026 inventory workbook. Its stage, rating and deployment date are read from the same dataset that powers the inventory dashboard. A blank means DHS left the field empty. Quoted text is exact. The chat windows are built from each filing's text, and the assistants did not write them.

The STELLA layers come from ICE's September 2026 request for information and its Oct. 21, 2026 answers to vendor questions on SAM.gov. The inventory entry for ICE's assistant does not use the name STELLA. The link comes from FedScoop and Biometric Update reporting. The directive language comes from DHS Directive 139-08 Rev. 01 and OMB M-25-21.

Counts count what is listed on this page. They do not cover every generative AI use case in the inventory. The dashboard has the full dataset.

This document provides general information only and does not constitute legal advice. Immigration law is complex and individualized. Outcomes depend on specific facts and circumstances. No results are guaranteed. Always consult with a qualified immigration attorney before making decisions about your immigration case.