Immigration Agencies Operating Dozens of Undisclosed AI Systems, Raising Executive Governance Concerns

Summary

U.S. immigration agencies have deployed approximately 117 AI systems across Immigration and Customs Enforcement (ICE), Customs and Border Protection (CBP), and the U.S. Citizenship and Immigration Services (USCIS), but significant gaps exist between actual deployments and public disclosure. FOIA litigation by civil rights organizations revealed that ICE operated roughly 42 AI tools as of May 2024, with only 23 appearing in DHS's public AI Use Case Inventory. CBP similarly operated approximately 75 AI systems, though that number has since been reconciled with public records. This transparency gap highlights a broader governance challenge in how AI is reshaping executive power without adequate public oversight.

The deployed systems include tools such as ImmigrationOS, which combines government and commercial data to prioritize enforcement targets; Hurricane Score, an algorithm that predicts whether immigration program participants will abscond using demographic factors that raise equal protection concerns; and USCIS Asylum Text Analytics, which flags asylum applications for additional scrutiny and monitors attorneys as a network. These systems operate at speeds that can outpace meaningful human oversight and complicate accountability when decisions go wrong.

The deployment pattern mirrors historical precedent: post-9/11 surveillance infrastructure developed for immigration enforcement—including watchlisting systems, biometric databases, and information-sharing networks—eventually expanded to broader domestic surveillance affecting all Americans. Scholars argue that immigration now serves as a 'test case' for how AI could erode constitutional protections and the separation of powers, given that noncitizens face weaker constitutional protections and more deferential judicial review. The concept of "ExecAI" describes how advanced systems in the executive branch could execute presidential directives with minimal institutional friction, reducing opportunities for dissent, whistleblowing, meaningful attribution when errors occur, or time for courts to intervene.

• Historical patterns suggest surveillance and decision-making infrastructure first normalized on noncitizens eventually expands to broader populations, raising concerns about long-term effects on all Americans
• FOIA litigation by civil rights organizations is bringing previously undisclosed AI systems to light, but regulatory frameworks for AI governance in executive agencies remain underdeveloped