TL;DR: Immigration enforcement in the United States has entered a new phase of automation. Nearly a year after US Immigration and Customs Enforcement began rolling out ImmigrationOS, an artificial-intelligence platform built by Palantir Technologies, the system has begun quietly reshaping how the agency identifies and tracks people marked for deportation.
What originated as a narrow procurement notice has grown into a broader effort that links machine-learning models with government and commercial records, creating an enforcement infrastructure with few historical parallels. As the software moves from early pilot to wider operational use, the arguments around it have hardened.
Supporters inside the government describe it as a long-overdue modernization of a fragmented, manual system. At the same time, critics in civil-rights and legal circles see it as a test case for how deeply AI systems can reach into civilian life before courts, lawmakers, and technical standards catch up.
In early 2025, ICE requested what it called a "streamlined end-to-end immigration lifecycle" platform and awarded a multimillion-dollar contract to Palantir to build ImmigrationOS, with a prototype due by September that year. Palantir designed the system to prioritize individuals for removal – ranging from visa overstays to violent offenders – by consolidating data that previously resided in separate systems, including criminal records and civil immigration files.
Technically, ImmigrationOS sits on top of Palantir's existing data-integration stack, originally built for military and intelligence work and later adapted for domestic law enforcement. These platforms ingest structured and unstructured data from dozens of sources, normalize identifiers such as names and addresses, and expose them through search and analytic tools that can generate what ICE agents call "targeting packages."
The new system extends that model more aggressively into immigration: algorithms are tasked with reconciling identities across incomplete or noisy data sets, surfacing anomalous immigration histories, and ranking leads for field officers. For years, ICE has been expanding the raw material that those models can draw from, pulling records from state motor vehicle departments and the Social Security Administration. It can also access local law-enforcement databases, jail and court systems, and commercial aggregators that assemble information on utilities, phones, and financial activity.
The immigration agency's data architecture combines traditional records with commercially available and app-derived information. Purchases from advertising-technology brokers add location and behavioral data linked to smartphones and apps, while automated license-plate readers contribute detailed travel histories over time. The agency also draws on video and biometric tools: it can request footage from more than 2,000 local police and fire departments that partner with Amazon's Ring.
Homeland Security Investigations uses specialized vendor tools to identify and track people of interest. Meanwhile, ICE holds a facial-recognition contract with Clearview AI, whose database includes tens of billions of images scraped from the public internet, which it uses to search for matches on priority suspects. Another vendor, under a confidential agreement, provides object-matching capabilities, identifying recurring cars or clothing across multiple videos by tracking distinctive visual features such as damaged bodywork or tears in fabric.
On the software side, one of the most immediate uses for automation is paperwork. Preparing affidavits and other documentation for subpoenas and warrants has traditionally taken investigators days of manual work, as they navigated between disconnected systems. By automating large parts of that process, AI tools can now generate drafts in under an hour in many cases, a change expected to increase the volume of judicial requests, even though each still requires human review.
Immigration officials are configuring ImmigrationOS to alert them when potentially relevant data exists behind privacy or policy walls and to recommend legal mechanisms – such as specific types of court orders – that agents can use to obtain it. Despite this technical sophistication, the reliability and governance of these tools remain unclear. Civil-rights researchers argue that error rates are still too opaque and that mistakes tend to fall on people with common names or dense digital footprints.
Courts and regulators are beginning to delineate boundaries. A federal judge ruled this month that the Internal Revenue Service had illegally shared taxpayer data with the Department of Homeland Security, banning further access and signaling that some forms of inter-agency data flow exceed existing legal authority. Several state and local governments have also moved to restrict voluntary data sharing for civil immigration enforcement, particularly in jurisdictions that have adopted sanctuary policies, even as those same records may end up with ICE indirectly through commercial resellers.
Those indirect channels are now central to ICE's data strategy. In Cook County, Illinois, where local policy limits cooperation on civil immigration cases, jail records can still reach ICE after being sold to companies like LexisNexis Risk Solutions, which then licenses access back to federal agencies. Lawyers working on these issues say the contracts can be complex enough that local officials may not fully realize how and where their data travels once it enters the commercial market.
At the same time, ICE's own budget for information technology and data services has expanded sharply, with DHS awarding more than $1 billion in IT contracts over roughly the first year of President Trump's second term, including tens of millions earmarked for Palantir's work on ImmigrationOS. The influx of funding has allowed the agency to scale its data collection, integration, and analysis capabilities far beyond what was possible in previous years.
The wider surveillance apparatus has begun to reach into political activity as well. Reports from advocacy organizations and legal clinics describe ICE compiling information on activists who attempt to disrupt enforcement actions, using the same imaging and social-media tools that underpin other parts of the agency's work. Those efforts, they say, have also swept in data on lawful protesters and observers, raising concerns that the presence of cameras, facial-recognition systems, and AI-driven flagging tools could chill participation in demonstrations.
For technologists, ICE is no longer simply consuming off-the-shelf databases; it is orchestrating a layered stack of analytics, from identity resolution to pattern detection to workflow automation, all tuned to decisions about who gets investigated, arrested, or removed. It remains to be seen whether that stack ultimately becomes a standard model for other agencies – or is curtailed by new rules, court decisions, and public pressure.