Phone calls, messages, and video visits from people in immigration detention could soon feed a single AI-powered system that builds voice profiles, interprets slang, and maps whom detainees talk to and spend time with, according to a request for information the federal government posted Friday.
U.S. Immigration and Customs Enforcement posted the request on SAM.gov, the federal contracting website. ICE facilities produce phone, message, video, and camera data “at volumes that cannot be reviewed and correlated manually in a timely manner,” according to the accompanying draft work statement.
The work statement describes a surveillance system for detention facilities. It would combine and analyze what detainees say, how their voices sound, where they go, and who they spend time with.
Requests for information are not formal solicitations or requests for proposals. Agencies typically use them as a first step to gauge whether industry can deliver a capability, and they do not commit the government to issuing a solicitation or awarding a contract. The requirements in the work statement could change or never lead to a purchase.
A shared database would automatically link calls, messages, voice matches, camera events, housing assignments, movement records, and investigative files. It would connect to “investigative or case-management sources designated by the Government,” which the work statement does not name. ICE’s existing case management system is called Investigative Case Management and is supplied by Palantir.
Analysts could view a detainee’s contacts and activities as a relationship map or timeline. Vendors may also offer “outside-party identity enrichment” from public or commercial data sources, which ICE would have to specifically authorize before use.
Voice profiles, or digital records of how a person’s voice sounds, would be built from calls ICE has already recorded. Detainees would not be asked to give a voice sample or to cooperate in any way, and ICE would not need new recording equipment, according to the work statement.
A call would be flagged if the caller’s voice does not match the person who holds the account, which could mean someone is using another person’s account or PIN. The system could also compare whoever answers a call against a voice profile tied to the number that was dialed. That would bring relatives and others outside the facilities into the analysis. The same voice could be matched across accounts, phones, and facilities, and, with separate approval, across other agencies.
Software would scan calls and messages for what the work statement calls “coded language,” slang, and words or phrases that start turning up more often than usual. Investigators would get summaries and alerts.
Cameras would be watched for fights and for behavior the work statement labels “isolation or withdrawal,” “unusual congregation,” and “unusual movement or interaction.” Staff would also be alerted to signs of distress or self-harm, and analysts could track who interacts with whom over time.
A human being must review the system’s findings, and the work statement says anything it produces cannot be the only basis for enforcement or punishment. ICE’s standard contract language, approved by its legal office on July 16, 2025, more than a year before this request, also bars using the technology to predict a person’s “future behavior or emotional state” in ways that lead to discriminatory or unlawful actions. The work statement says that language will be added to all ICE contracts.
Nothing in the text explains how a detainee or family member would find out about a wrong voice match or challenge it. It also does not say whether signs of distress flagged for staff could be used for security or discipline.
Monitoring detainee calls is nothing new. Under ICE’s detention rules, a facility can monitor calls for security reasons at its own discretion. It must have a written policy and tell detainees about the monitoring in its handbook and on a notice posted at each monitored phone. Phone and video calls about legal matters cannot be electronically monitored without a court order. But the system described in the work statement would do much more than monitor calls. It would also handle messages, video visits, and camera footage, build voice profiles, and connect the results to detainees’ housing, movement, and investigation records.
The Federal Bureau of Prisons, part of the Justice Department, has already bought AI for monitoring communications. It selected Leo Technologies for a five-year contract worth more than $106 million, Correctional News reported in May. The company’s Verus platform will provide “AI-enabled translation and transcription services” for monitoring inmate phone calls and electronic messages and would search them for “actionable intelligence” on security threats, contraband, and gang activity. The report does not say whether the contract includes voice identification or video.
DHS keeps a public list of the artificial intelligence tools its agencies use. The list has 50 entries for ICE, and this system is not among them. Two earlier ICE tools, one that analyzed voices and one that analyzed video, are listed as discontinued.
A White House Office of Management and Budget memo from April 2025 requires extra safeguards for federal AI that could affect people’s rights or safety, which it calls “high-impact.” The safeguards include testing before use, ongoing monitoring, and a way for people to appeal.
That label applies when an AI system’s output is the “principal basis” for a decision with a significant effect on a person. ICE has reviewed 15 of its tools that were presumed to be high-impact and decided none of them were, mostly because a person checks the results before anyone acts on them.
The work statement posted today requires the extra safeguards only if this system is eventually labeled high-impact.
Throughout, the work statement refers to the technology as “SI,” for “super intelligence,” the term President Donald Trump ordered federal agencies to adopt in an executive order last month.
ICE did not respond to a list of questions that Project Salt Box emailed Friday.
Of course! Mullin and Vought must love to have ICE/DHS out there on the cutting-edge of neofascist AI development and deployment.