OP-ED: The $100 Million Hole in ICE's Deportation Machine

A procurement notice describes the one capability ICE doesn't have yet: a standing near-real-time intelligence enterprise to turn its surveillance into target lists.

(Photo: Chris Yang via Unsplash)

The morning of September 11, 2001 needs no introduction. The events of that day fundamentally ruptured the workings of the world as we knew it and ushered in an era defined by war and by a shrinking sense of safety that has never fully returned. From where I sat in homeroom, watching the towers smolder on the live TV Mrs. Jones wheeled into the room, I couldn’t fathom how. But the world had changed.

In the intervening years, the vector of the American psyche bent towards revenge. The campaigns that made up the so-called Global War on Terror killed more than 940,000 people directly in Iraq, Afghanistan, Syria, Yemen and Pakistan between 2001 and 2023 — more than 432,000 of them civilians, according to the Costs of War project at Brown University. An estimated 3.6 to 3.8 million more died indirectly, from the destruction of economies, health systems and infrastructure, bringing the total to at least 4.5 million.

On the home front, the Patriot Act forged new mechanisms of surveillance — a “naked power grab by the government itself,” according to the bill’s lone opponent in congress, Senator Russ Feingold of Wisconsin. But America was licking its wounds and wanted its justice.

At the same time the Department of Homeland Security was opening its doors in 2003, I was going through the process to enlist in the Air Force. I wanted to be a crew chief. My recruiter asked me to take some additional tests, and I scored high enough on a battery of strange examinations to render myself ineligible for menial work. “Have you ever considered foreign languages?” he asked.

At my high school graduation, when they read the list of names of people who were going to serve in the armed forces, nearly a quarter of my graduating class stood up.

I served as a cryptologic language analyst. My job was to translate and analyze foreign communications and turn what I found into intelligence that decision-makers could act on. I did it for close to twenty years. This is not a story about my time in service. It is a story about how, 25 years to the day after the fabric of the world was ripped apart and hastily sewn back together, the Department of Homeland Security advertised its desire to build an intelligence apparatus whose description closely mirrors the work I spent my career doing.

I recognized it on the first read.

The advertisement, titled “ERO OVERWATCH AND NATIONAL SECURITY ENFORCEMENT (ERO ONE) — INTELLIGENCE RESEARCH SPECIALISTS PROFESSIONAL SUPPORT SERVICES,” was posted to DHS’ Acquisition Forecast Planning System — a system the department uses to flag upcoming contracting opportunities before beginning the formal bidding process. In short, it’s a heads up to contractors.

I used to joke with a colleague that nothing in the military gets done until it has a name. Ideas circulate and die. Give the same idea an acronym and it acquires a budget, a briefing slide, a program manager, and a constituency of people whose careers depend on it. A name is a powerful thing. It is often the mechanism by which a mere proposal becomes a thing that exists. Older traditions understood this better than we do. Grimoires, the manuals of ceremonial magic that circulated from the medieval period onward, were organized substantially around names: to summon a spirit you needed its correct name, and to command it you needed to speak that name precisely, on the theory that a thing properly named must answer.

Governmental bureaucracies run on a version of this. A thing without a name is a suggestion that can be deferred or stolen or quietly killed by a budget officer. A thing with a name can be scheduled, funded, staffed, and reported on. Providing direct intelligence support to ERO is an idea. ERO ONE is more than that. It implies byzantine governmental structures, objectives, desired outcomes, iteration, and results. That ERO ONE has a name is telling.

The government expects to spend more than $100 million on the contract. It plans to release the formal solicitation by Dec. 10 of this year, make an award sometime in the spring of 2027, and keep the winning contractor on the job for five more years after that, through February 2032.

To understand what that means, you have to understand the sources of information already at ICE’s disposal. The agency is not a member of the intelligence community, so it doesn’t have access to the classified troves that membership would bring, though it has lobbied to join for years. But it already runs a wide network of law enforcement data tracking people inside the United States. License plate readers. Biometric scanners. Port of entry records. State motor vehicle files. Federal criminal databases including the National Crime Information Center, the National Instant Criminal Background Check System, and the Combined DNA Index System.

Those feeds pour into shared government repositories: the Automated Biometric Identification System, known as IDENT, which holds records on more than 270 million people and is being folded into a successor called the Homeland Advanced Recognition Technology system, or HART; the Treasury Enforcement Communications System, or TECS, which logs border crossings and encounters; and the Enforcement Integrated Database, or EID, which tracks arrests, detentions and removals.

From there the information flows into a platform ICE Homeland Security Investigations calls RAVEn, short for Repository for Analytics in a Virtualized Environment, which was built to let analysts search across all of it at once. RAVEn pulls from the private sector as well. Clearview AI, which built its face-matching database by scraping images off the open internet, holds a $3.75 million contract with the agency. Penlink, a Lincoln, Neb., company that acquired the Israeli firm Cobwebs Technologies in 2023, sells ICE two tools: Tangles, which scrapes the open, deep and dark web and can add facial recognition as a module, and Webloc, which tracks people’s locations using data harvested from mobile phone advertising. Paragon Solutions sells Graphite, spyware capable of pulling data directly off a target’s phone. Contractors including Global Recovery Group perform skip tracing, the practice of locating people who have moved or gone off the grid.

In the field, agents carry an app called Mobile Fortify, which lets them point a government-issued phone at a person on the street and run their face and fingerprints against those federal repositories in real time. ICE has used it more than 100,000 times since launch. And a Palantir-built case management system called ImmigrationOS pulls the finished picture together for the agents making the arrest.

Now, imagine a layer between all of this technology and an ICE agent in the field. That layer is what ERO ONE is likely meant to be. It would be a team of analysts dedicated to making sense of that web of data and turning it into actionable intelligence to drive ICE enforcement and removal operations on American streets.

The listing is vague about the particulars, as forecast notices tend to be, but the language it uses to describe the work is lifted straight from the intelligence community: all-source analysis, network and relationship analysis, intelligence gap analysis, intelligence product development, field intelligence support. This could easily be the job description of an intelligence analyst supporting military operations, and that arrangement is worth understanding in some detail, because it is the closest thing we have to an analog for what ERO is proposing to build.

For most of the last twenty-five years, in Afghanistan and Iraq and later in Syria and Yemen, the work of finding a person — a target of interest — was distributed across the intelligence community and then concentrated into singular assessments. The National Security Agency collected and exploited foreign signals intelligence. The National Geospatial-Intelligence Agency delivered geospatial intelligence, the exploitation and analysis of imagery to describe and depict physical features and activity on the earth. The Defense Intelligence Agency provided all-source defense intelligence to warfighters and policymakers, including human intelligence collected by its case officers. Beneath them, the military intelligence elements of each service did the same work closer to the ground, each responsible for supporting its own commanders. None of those streams was sufficient by itself, and none of them agreed cleanly with the others. The analyst’s job was to sift through all of it, weigh the reliability of each source against the rest, and produce a judgment a commander could act on: this is the target, this is where he is, this is the confidence we have in saying so. That judgment was briefed to the unit that carried out the operation, and whatever the unit recovered went back to the analysts to inform the next one. Operations generated intelligence, intelligence generated operations, and the cycle sustained itself for two decades.

I want to be precise about how that judgment actually worked, because the word intelligence suggests a confidence the job rarely had. Numerous were the times our target validations came back incomplete and a strike or an arrest could not happen. The package did not meet the standard, so the mission did not go. Less often, an incomplete package went forward anyway and the mission was prosecuted. Bear that in mind. There were layers of redundancy, review at multiple levels, people whose entire function was to find the flaw in someone else’s work, and mistakes still happened. The system was built on the assumption that analysts would be wrong sometimes, and nearly all of its machinery existed to catch that before anyone acted on it.

ERO ONE describes the same function, rebuilt at a smaller scale and pointed in the opposite direction. The collection agencies are replaced by the commercial and law enforcement systems already in ICE’s hands. The commander is a field office director. The unit that acts on the assessment is a removal team, and the person being fixed is not in Kandahar. He is in Baltimore, or Chicago, or a town in central Iowa, or Memphis.

The kind of work ERO ONE resembles was bound by safeguards that have no distinguishable counterpart here. United States persons were off limits. The rule was a body of law and procedure that governed what could be collected, what had to be minimized and what had to be purged. Compliance was rigorously verified rather than assumed. By directive, the National Security Agency conducted site visits to examine whether the safeguards at a given location were actually operating as intended, ran sampling reviews of analyst queries against raw traffic databases to confirm they complied with the rules protecting U.S. persons, and independently tested purge processes to verify that material required to be destroyed had in fact been destroyed. Behind that sat trained auditors, executive orders, two congressional oversight committees, inspectors general, and a court. The system was imperfect and was violated, sometimes badly, and the violations that became public are the reason several of those controls exist in their present form. The controls were prohibitions, analysts were trained in them before they were trained in almost anything else, and the whole edifice rested on a single distinction: a respect for the constitution and the law.

That distinction was enforced by more than bureaucracy. My colleague Daniel Hulter, who I served alongside for years and who retired this summer after twenty years and four months in uniform, wrote recently about what the enforcement of it looked like from the inside. On his first day on the joint operations center watch floor at Bagram Air Base, Afghanistan, in 2011, a strike went in. At the moment of impact the room erupted in cheers. He wrote that his stomach sank. Weeks later, under a new commander, the room erupted again, and the commander shut it down on the spot, making clear that killing would not be celebrated on his watch. Hulter said he nearly cried from relief that someone with authority had taken the room seriously.

I recognize the room he is describing. By the time I was senior enough, the services were building that recognition into military education curricula rather than leaving it to whichever commander happened to walk in. In 2022 I helped write the requirements that put moral injury into career field level training for all intelligence career fields, on the premise that people who do this work will be marked by it and that an institution has an obligation to prepare them for that. The effect of teaching it is not only that analysts are better protected. It is that they do the work more carefully. An analyst who has been taught that a bad call will stay with him treats the call differently, and that is the closest thing to a guardrail that operates when nobody is watching. A statute constrains what you are permitted to do. A person with authority constrains what happens in the room. Training is what constrains you at three in the morning, alone with a partial identity and a deadline. The military spent twenty years learning it had to teach that, and built the requirement into the career field. There is no equivalent on the other side. A contractor hired into ERO ONE would arrive with a clearance and a queue.

DHS has something closer to a direct analog. The Office of Intelligence and Analysis maintains Intelligence Oversight Guidelines that cover information about Americans and require an oversight officer to run spot checks, review audit logs, and interview contractors. On paper it is the same architecture I worked under.

In practice it is not. The Government Accountability Office found in 2023 that the office was skipping two audits its own rules require. Asked why nobody was auditing bulk data, the council responsible said it was not their job. Over five and a half years, the office completed one compliance review. The rules I worked under were checked continuously. These were checked once.

And they may not apply here at all. The guidelines govern DHS intelligence components. ERO is not one. What this contract describes is law enforcement work at an agency outside the intelligence community, which puts it outside the reach of the only rulebook DHS has.

The difference is in what those instruments do. The intelligence regime tells an analyst what they may not collect and then audits them to confirm they did not. The DHS regime tells a component to document what it is collecting and publish a notice about it. The first is a prohibition enforced by inspection. The second is a disclosure requirement enforced by paperwork, and the paperwork is graded by the same department that produced it. When ICE fielded Mobile Fortify, its own initial review found the app privacy sensitive and in need of a detailed public report. The agency then determined that a privacy assessment written in 2019, for a different system, would do.

There is also the matter of who the protections cover. The Privacy Act extends statutory rights to citizens and lawful permanent residents and does not reach visitors or noncitizens. DHS closed that gap by policy in 2007, declaring that it would treat personal information the same regardless of immigration status, partly on the reasoning that the department could not credibly ask foreign governments to protect Americans’ data while refusing to protect theirs. In 2017 the department rescinded that policy to comply with an executive order directing agencies to stop extending Privacy Act protections to anyone who is not a citizen or a green card holder. What remains for everyone else is a set of fair information practice principles, which is to say a statement of good intentions that a subsequent executive order can revise again.

That is the oversight environment into which ERO ONE would be delivered. Rules exist. They are built to document rather than to forbid, they cover the smaller half of the population the system will examine, and they are administered by the department doing the examining.

The objection Russ Feingold raised in October 2001 was not that the government would abuse its new authorities immediately. It was that authorities granted under the conditions of an emergency do not expire when the emergency does. The historical record has borne his prediction out. Section 213 of the Patriot Act authorized delayed-notice search warrants, known as “sneak and peek warrants,” on the argument that agents needed to search a suspected terrorist’s home without tipping him off. Of the 3,970 such warrants requested in the year ending September 2010, more than 76 percent were narcotics cases and 37, less than 1 percent, were terrorism cases. By fiscal year 2020, courts were issuing close to 20,000 of them annually, with drug cases accounting for more than 70 percent. The authority did not shrink when the justification for it faded. It found new work.

Earlier this year, Elizabeth Beavers, a law professor at Widener University, documented five legal precedents from the war on terror that made this possible. The first is the conflation of immigration enforcement with counterterrorism, and the pattern she traces is consistent enough to be a method. In the days after the attacks, the FBI ran an investigation that leaned on immigration law rather than criminal law and swept up roughly 1,200 Arab, Muslim and South Asian men into secretive detention. Many were held for weeks or months without charges. The program produced no convictions for terrorism, and what it did produce were hundreds of closed-door proceedings over technical immigration violations, including students carrying too few academic credits. A registration program launched the following year, authorized in part by the Patriot Act, required noncitizens from 24 Muslim-majority countries and North Korea to register, submit biometric data and subject themselves to interrogation and tracking. It convicted no one of terrorism and deported more than 13,000 people, most for paperwork violations.

Beavers argues that the current administration extended the precedent to its logical conclusion by framing migration itself as terrorism, and the framing is not incidental. The president has called loose immigration policy the single greatest national security threat facing the country. The National Security Strategy released last November declared that border security is the primary element of national security and listed unchecked migration first among examples of invasion. Day-one executive orders declared that people present unlawfully are involved in preparations for terror-related activities, and directed Northern Command, itself a creation of the post-9/11 reorganization, to develop a campaign for repelling unlawful mass migration. In December, the director of the National Counterterrorism Center told the House Homeland Security Committee that a review of immigration cases from 19 countries had already identified about 18,000 known and suspected terrorists inside the United States subject to deportation.

Here we see the same mechanism that produced the Patriot Act and DHS in the first place. An event is designated a threat to national survival, the designation justifies authorities that would not otherwise pass, and the authorities outlast the designation. The administration has framed immigration as an invasion, a term chosen for its legal weight as much as its rhetorical effect, and that framing is what converts civil immigration administration into a national security mission requiring military equipment and methods. ERO ONE is another step in transforming federal law enforcement into a nationwide military campaign.

The operational half of the war on terror has already come home, and has been home for some time. ICE was built in 2003 out of the fusion of immigration enforcement with counterterrorism, given formal ties to the FBI’s Joint Terrorism Task Force and a mandate framed as national security rather than civil immigration administration.

In the past two years, the resemblance has stopped being merely structural and become cultural and operational. Federal procurement records reviewed by The Lever show ICE contracting with private security firms, several of which had never held a federal contract, to provide firearms and combat training to its snipers and special response teams. Multiple federal agencies raided a Chicago apartment building with a Blackhawk helicopter and flash-bang grenades, on public assertions that members of a designated terrorist organization lived there. None of the 37 people arrested was ever charged with a crime. Arrest records later obtained by ProPublica show agents entered with the building owner’s written consent, described the operation as based on intelligence that people were occupying apartments unlawfully, and never mentioned the gang at all.

Grant Tocsin, writing in the Small Wars Journal, argued that federal immigration agents have adopted the aesthetics and tactics of special operations forces with minimal operational justification, and that the agency’s recruiting appeals to a warrior identity rather than a law enforcement one. Meanwhile, Julian Go — a sociologist at the University of Chicago — calls this the imperial boomerang, in which methods developed for use on populations abroad ultimately return to be used on populations at home. And Spencer Ackerman, whose book Reign of Terror documents how DHS used counterterrorism language to target immigrants from its earliest days, summarized that trajectory plainly last year: ICE will perform mass extraordinary renditions at scale.

The analytical layer proposed in the ERO ONE forecast is what has been missing. Tactical teams, armored vehicles, helicopters and biometric scanners belong to the finishing end of the cycle. They are what happens after someone has decided who to visit and where to find him. ERO ONE is the other end, the part that produces the decision, and it is the part that made the difference overseas. The raids in Iraq and Afghanistan were not effective because the assault elements were skilled, though they were. They were effective because a standing analytical enterprise turned every completed operation into the starting point for the next one, continuously, for twenty years. That is the capability DHS has now advertised an interest in buying, on a six-year contract, pointed at American neighborhoods.

The listing specifies contractor personnel rather than federal employees. The contract would be awarded under OASIS+, a governmentwide vehicle whose unrestricted pool for this category holds 674 eligible vendors, among them Booz Allen Hamilton, CACI, Leidos, ManTech, Peraton, General Dynamics Information Technology, and The Mission Essential Group, which built its business supplying linguists to operations in Afghanistan. A contract analyst answers to a company before an agency. They have no civil service protection, no independent channel to an inspector general, and a direct professional interest in producing the volume of finished product the contract measures them against.

ICE has already run this play at smaller scale. Last fall, the agency put out a request for information seeking nearly 30 contractors to staff around-the-clock social media monitoring at two targeting centers, one in Williston, Vt., the other in Santa Ana, Calif. Senior analysts, shift leads and researchers, working in rotation, combing Facebook and TikTok and Instagram and turning what they found into enforcement leads for ERO. The documents set production deadlines as short as 30 minutes for a high-priority case.

An analyst supporting an operation in Afghanistan who assessed a target as too uncertain to act on was doing the job correctly, and a bad strike drew an investigation. The use of force in those operations was constrained by the Law of Armed Conflict, by rules of engagement written for each theater, by the Geneva Conventions and the treaty obligations the United States has ratified, and by a military justice system that could prosecute violations.

None of that framework applies to a removal operation in Minneapolis, and what replaces it is thinner than most people realize. The warrants ICE issues are largely administrative, signed by an agency officer rather than a judge. DHS rated Mobile Fortify high impact while its monitoring and appeals processes were still unfinished. And the machinery for investigating agents has been dismantled from the inside. In March 2025 the department all but eliminated its Office for Civil Rights and Civil Liberties, the office that had handled use-of-force complaints, abandoning 550 pending cases. Acting Director Todd Lyons told the Senate in February that ICE had opened 37 use-of-force investigations over the preceding year across a workforce that had nearly doubled to 22,000. He did not say whether any had produced a termination.

The agent who shot Renee Good was placed on administrative leave for three days and then quietly transferred out of state. When reports placed him in New York, Gov. Kathy Hochul wrote to the White House border czar and the homeland security secretary demanding to know whether it was true, saying reassignment is not accountability. Three months earlier, an officer filmed shoving a detained man’s wife to the ground in New York was placed on leave and back at his job within 72 hours.

Under the Trump administration, ERO operates under numeric expectations. In that environment, the analyst who resolves an identity and produces an arrest is performing. The analyst who counsels caution is slowing the line.

ICE will surely say there is a legitimate case for ERO ONE. The agency has people with final orders of removal who have not left, finding them takes work, and analytic support makes that work faster. Faster is the whole argument, and it does not survive contact with what ICE already has. The agency runs facial recognition on the street. It pulls from license plate readers and state motor vehicle files, scrapes social media, has purchased spyware that can read a phone directly, pays digital bounty hunters to trace people who have moved, and runs a Palantir system that assembles all of it into a case file. What is missing is not information. A dedicated, around the clock targeting enterprise on top of that stack is not a fix for a gap in capability. It is overkill.

The numbers demonstrate what the extra capability would be used for. ICE reported 356,389 removals in fiscal 2026 through July 11, already past the 319,980 it recorded for all of fiscal 2025, at roughly 1,250 people a day. Of the 65,765 people in detention that week, 70.6 percent had no criminal conviction. The agency prefers to say 60 percent have a criminal record, but it gets there by counting pending charges as convictions. Fewer than three in ten had actually been convicted of anything, and Cato put the share with violent convictions at 5 percent. Ninety-two percent of the growth in detention in fiscal 2026 came from people with no charges or convictions at all.

The Marshall Project, analyzing ICE detention records obtained through the Deportation Data Project, documented how rapidly people are moved between facilities and across state lines, a pattern that separates them from attorneys and families at exactly the moment representation matters most. OPB found detainees put on as many as six flights and flown thousands of miles, missing court dates in the process. This is not an agency struggling to locate the people it is looking for. It is an agency processing as many people as it can, as fast as it can, and ERO ONE would make the front end of that faster.

The consequences are already documented. Rep. Bennie Thompson of Mississippi, the ranking Democrat on the House Homeland Security Committee, told 404 Media that ICE treats a Mobile Fortify match as a definitive determination of status, and that an officer may disregard a birth certificate if the app returns a different answer. Jesus Gutiérrez, a U.S. citizen walking home from the gym in Chicago, was handcuffed and held for an hour while agents scanned his face, according to a complaint filed by Illinois and the City of Chicago. He had tried to show them his documents on his phone. ProPublica has documented more than 170 U.S. citizens detained by immigration agents in nine months of counting.

None of those people were the intended targets. They are what a system built for speed and scale produces, which is why this should concern citizens and non-citizens alike. An analyst resolving an identity works from the same license plate readers, the same phone location data, the same scraped social media accounts, regardless of who is on the other end. Citizenship gets determined at the end of that process, after the surveillance and often after the detention. A system built to find one category of person inside the United States has to examine everyone in order to sort them.

The same forecast system that DHS is using to inform potential contractors about ERO ONE is the one it used before spending $16.7 million on 6,000 pairs of electric shock gloves last month. The agency posted that requirement on Aug. 10, issued the solicitation four days later, and awarded a no-bid contract soon after. Those gloves, by the way, deliver a shock of more than 300 volts on contact. Three members of the House Homeland Security Committee wrote to Secretary Markwayne Mullin asking him to stop the purchase. Sixteen senators wrote separately asking about testing, oversight and limits on use. The contract was awarded anyway.

ERO ONE will not horrify anyone at first glance the way shock gloves do. It should. And the reason it should is that a version of it is already running. This week 404 Media revealed the existence of Border Patrol units called Predictive Intelligence Targeting Teams, which analyze Americans’ financial records and pass what they find to local police. In Montana, a PITT reviewed a man’s transactions, flagged patterns it associated with narcotics activity, and local officers pulled him over for an obstructed license plate. That is parallel construction: the stop is given a lawful-looking reason so the analysis that produced it never has to be disclosed or defended. 404 Media found the teams in the Spokane and Laredo sectors. CBP runs twenty sectors and declined to say how many more exist. Rob Frommer of the Institute for Justice called the combination of mass surveillance and predictive policing a recipe for tyranny.

One could argue the opposite of everything I have written here. More human judgment in the targeting process is better than less, and an analyst who slows an operation down is exactly the safeguard a system of automated matches and administrative warrants currently lacks. I would find that persuasive if ERO ONE were being built as a check. It is not. A validation cell sits outside the operational chain and can stop it. What this forecast describes sits upstream of the arrest and is measured on what it produces. Under an agency that treats a face scan as conclusive, moves people beyond the reach of counsel, and has detained citizens by the hundreds, adding analysts does not add restraint. It adds capacity. Ask the man in Montana.

The danger of ERO ONE is therefore not contained in the contract itself. It lies in what happens when a quarter-century of surveillance authorities, databases, commercial data, biometric systems, and military-derived intelligence practices are brought together and given an institutional mandate to find people. Each component was built at a different moment, under a different rationale, and with a different set of constraints. But as a whole, they constitute something much larger: a system capable of identifying, locating and directing the state’s coercive power toward individuals across the country. ERO ONE would give that system a more permanent analytical function. Once that capability exists, the question will no longer be what information the government possesses. It will be what it chooses to do with it.