ICE seeks AI surveillance system to link detainees’ calls, contacts and movements

U.S. Immigration and Customs Enforcement is asking industry about an AI system that would connect detainees’ conversations, voice identities and movements inside detention facilities. ICE posted the request on SAM.gov on October 9, according to Project Salt Box. It is a nonbinding inquiry—not a contract award—but outlines surveillance reaching beyond individual calls.

The accompanying draft work statement says detention facilities generate more phone, message, video and camera data than staff can manually review and connect quickly. Its proposed answer is a shared database linking communications to voice matches, camera events, housing assignments, movement records and investigative files.

Requests for information let agencies gauge what industry can deliver. This one does not commit ICE to a formal solicitation or purchase, and its requirements could change. The document describes desired capabilities, not a system already operating in detention facilities.

ICE’s detention rules already allow facilities to monitor calls for security reasons. Facilities must have a written policy and notify detainees through their handbook and notices at monitored phones. Phone and video calls about legal matters cannot be electronically monitored without a court order.

The proposed system would go beyond listening to a conversation. Analysts could see a detainee’s contacts and activities as a relationship map or timeline. It would connect to government-designated investigative or case-management sources, although the draft does not name them. Vendors could also offer identity information about outside parties from public or commercial sources, subject to ICE’s specific authorization.

What the draft would flag

• Communications: coded language, slang and words or phrases appearing more frequently than usual, with summaries and alerts for investigators.
• Camera footage: fights, isolation or withdrawal, unusual gatherings, and unusual movement or interaction.
• Welfare concerns: signs of distress or self-harm that would trigger staff alerts. Analysts could also track interactions over time.

Voice profiles—digital records of how someone’s voice sounds—would be built from calls ICE has already recorded. The draft says detainees would not need to supply a sample or cooperate, and ICE would not need new recording equipment.

A call would be flagged when the caller’s voice does not match the account holder’s profile, potentially indicating use of someone else’s account or PIN. The system could also compare the person answering against a voice profile associated with the dialed number, bringing relatives and other outside contacts into the analysis.

The same voice could be matched across accounts, phones and facilities, and across agencies with separate approval. The draft does not explain how a detainee or family member could learn about or challenge an incorrect voice match.

The draft requires a person to review findings and prohibits using the system’s output as the sole basis for enforcement or punishment. It also incorporates ICE contract language barring predictions of future behavior or emotional state that lead to discriminatory or unlawful actions. It does not specify whether distress alerts could be used for security or discipline.

Additional safeguards depend on whether the system is designated high-impact. An April 2025 Office of Management and Budget memo requires testing before use, ongoing monitoring and an appeal route for qualifying federal AI. The designation applies when AI output is the principal basis for a decision significantly affecting a person.

ICE previously reviewed 15 tools presumed to be high-impact and classified none that way, mostly because people checked results before action. For this proposed system, the draft makes the extra protections conditional on that designation. ICE did not respond to questions Project Salt Box emailed on October 9.