By Selen Ozturk
Contributing Writer
SAN FRANCISCO — At a high level, in order to effectuate their mass deportation agenda, the federal government is trying “to consolidate as much information as they can about individuals,” said Saira Hussain, a senior staff attorney at the Electronic Frontier Foundation. That surveillance apparatus pursues different avenues to gather data, she elaborated, including other federal agencies sharing data, as well as data obtained from states and local agencies administering federal grants.
This is the first in a two-part series exploring the government’s expanding use of surveillance technology as part of its immigration enforcement operations.
The same federal, state, local and private databases that immigration agents are using to identify and arrest immigrants could also violate U.S. citizens’ privacy, said lawyers, researchers and advocates at a Sept. 10 panel on federal surveillance held by American Community Media at the World Affairs Council.
What emerged is a consistent image of how the government’s deportation efforts deploy surveillance tools to gather data on immigrants in order to meet deportation and detention quotas set by the federal government. Between December 2024 and January 2026 alone, active immigration-related AI use cases by the federal Department of Homeland Security rose 36%.
Portland, Oregon-based immigration attorney Stephen Manning, who represents plaintiffs in a class-action case over a 2025 Department of Homeland Security surveillance operation, said enforcement effectiveness owes less to the sophistication of certain AI than its ability to compile disparate “streams of data” — sometimes before targets’ identities are even known.
He described his case, Operation Black Rose — an October 2025 DHS arrest of eight people in a van in Woodburn, Oregon, home to one of the state’s largest Latino farmworker communities — as “arrest first, justify later.”
While DHS agents identified the area as a “target-rich location,” they did not initially know the identities of the people in the van when they surrounded it, Manning said. They instead used various AI data systems — including Palantir’s ImmigrationOS deportation management platform and a facial recognition tool called Mobile Fortify — to “live target” the vehicle.
After stopping the van, agents used cellular facial recognition to scan and identify people they took out and handcuffed.
“At this moment in time, the government’s position,” Manning explained, “is that none of these people who are sitting on the sidewalk with arms behind their backs, handcuffed, are under arrest. No one is under arrest. That’s the position in court” — that it is a stop requiring only reasonable suspicion, rather than an arrest requiring probable cause.
Everyone in the van was eventually processed for deportation except the lead plaintiff, identified by her initials MJMA; Manning’s team intercepted her at an Immigration and Customs Enforcement (ICE) facility before she could be transferred out of Oregon, and she now represents a certified class.
A recent court order granting a motion to compel will now give Manning’s team greater latitude to investigate how federal surveillance tools were actually used to construct this “dragnet,” he said.
Asked how key such tools are to the cases he tracks more generally, he said, “It is the essence of what they do,” warning that the surveillance infrastructure will outlast this administration: “Once you build it, you just repurpose it. It doesn’t go away.”
For Steven Hubbard, a senior data scientist at the American Immigration Council, surveillance power “doesn’t necessarily come from a particularly sophisticated AI system. … it’s the enormous amount of data that government agencies can now search, connect and analyze.”
This data includes immigration, travel, biometric and commercial records; information increasingly drawn from federal, state and local governments; and data from brokers like Thomson Reuters and LexisNexis, which use a government privacy loophole to sell information that would otherwise require a warrant.
He grouped DHS tools into three categories from the agency’s own semiannual AI inventory: machine learning, “statistics on steroids” for fraud detection and national security risk scoring; language AI, used to translate, summarize and classify documents; and computer vision, which analyzes visuals including facial recognition and license-plate readers.
He likened the process to “skip tracing” — a technique long used by debt collectors and bail agencies in which information like addresses, phone numbers and property records are used to find someone’s whereabouts — but now at a technological scale increasingly difficult for human fact-checking to match.
That consolidation push has drawn legal challenges on multiple fronts.
A March 2025 executive order by President Trump directing agencies to eliminate “information silos” opened new ways for agencies to access data held by the Internal Revenue Service, Medicaid records, the U.S. Department of Agriculture’s SNAP database, the U.S. Department of Housing and Urban Development’s housing-verification system and state voter rolls.
One underscrutinized channel is the American Association of Motor Vehicle Administrators (AAMVA), a nonprofit operating a commercial driver’s-license database.
“Last year, there was a horrible accident that took place in Florida, involving a Punjabi Sikh driver, who was driving a big rig, who ended up crashing, and killed a few people, really horrific,” Hussain said.
The case drew attention to commercial licensing standards and immigration status. DHS allegedly stated that Harjinder Singh, the driver of the big rig, was undocumented, though Governor Gavin Newsom’s press office pushed back against the allegation, arguing that it was the Trump administration that approved Singh’s commercial driver’s license.
Hussain said that it was unclear what Singh’s immigration status is.
“To be clear, if you’re a commercial driver, you actually have to have status within the country,” she said.
Thereafter Florida sued California and Washington for issuing licenses to undocumented individuals. The Supreme Court denied the hearing. But last summer, DHS and the Federal Motor Carrier Safety Administration issued a demand to AAMVA and asked for its full records of all 17 million in their system. Twenty-two states, including California, sued in August to block the transfer, and a district court judge granted a temporary restraining order.
Hussain described United Auto Workers v. U.S. Department of State, a lawsuit filed by the Electronic Frontier Foundation with Muslim Advocates and Yale Law School’s Media Freedom and Information Access Clinic over a State Department program requiring some visa holders to keep their social media profiles public, potentially making personal online activity part of immigration vetting.
“This program really aims to identify and intimidate people who are engaged in disfavored speech,” Hussain said. “What is disfavored speech? It’s the government that decides it. People who have spoken out in favor of Palestinian rights, people who have commented on Charlie Kirk’s murder, people who have engaged in speech criticizing U.S. policy have all been identified as engaging in disfavored speech.”
DHS has separately issued hundreds of subpoenas to Meta, Google, Reddit and other tech companies to unmask anonymous critics of ICE operations, sometimes without adequately notifying affected users.
“The onus is not on the individual. It’s up to our policymakers to do things like pass comprehensive data privacy laws,” Hussain said. “The only way that we’re going to come back from this is if we scrap these systems altogether and make sure they cannot be resuscitated.”
Data-driven immigration enforcement operations are expanding into public spaces, added Denise Bell, human rights researcher and co-author of an American Civil Liberties Union report documenting civil rights violations by immigration enforcement agents in 2025.
Reviewing more than 1,200 enforcement incidents across eight states in 2025, using only public records, she said her team found 624 actions in “everyday places” like sidewalks, parking lots, bus stops, parks and wildlife reserves, plus 252 at places of employment like car washes, restaurants, rideshare lots and landscaping and construction sites, and 107 at courthouses and immigration check-ins.
Researchers also documented 432 incidents of misconduct, including use of force and 119 cases of retaliation against people for recording enforcement; and 437 incidents involving likely racial profiling.
Racial profiling, Bell explained, “was not incidental bias at the margin. It was core to how dragnet enforcement was operationalized in 2025. People were stopped and detained because of perceived race, ethnicity, spoken language, accent.”
The report ties the surge in individual misconduct not only to federal cooperation with state and local law enforcement — including agreements allowing local officers to participate in immigration enforcement — but also to eroding accountability, as agents increasingly work masked and in unmarked vehicles, resisting civilian attempts at documentation.
“What the administration is doing through its immigration enforcement agenda is creating a blueprint for authoritarianism,” Bell said. “The administration’s project in mass deportation is not just a direct instrument of power. It is about defining who belongs in the United States.”
Selen Ozturk writes for American Community Media.
