The numbers are almost too large to make sense of. In fiscal year 2026, Immigration and Customs Enforcement is spending $28.7 billion on surveillance technology. That figure is ten times the agency’s total surveillance expenditures over the previous thirteen years combined. Not per year — combined. In a single budget cycle, ICE has built the most expansive AI-driven surveillance apparatus in American law enforcement history. And it isn’t just targeting immigrants.
This is what that system looks like from the inside.
ImmigrationOS: The Platform at the Center
The operational core of ICE’s new surveillance posture is a platform called ImmigrationOS, built under a $30 million contract awarded to Palantir Technologies. The contract runs through September 2027, with a working prototype that was due by September 2025.
ImmigrationOS was designed around three explicit functions. First, to streamline the identification and apprehension of removal priorities — including people categorized as “violent criminals,” gang members, and visa overstays. Second, to track and report self-deportations with “near real-time visibility.” Third, to make the logistics of deportation more efficient through better identification at every stage of the removal process.
That second function deserves a close read. Near real-time tracking of self-deportations means building a system that knows, as it happens, when a person leaves the country without being formally removed — and attributes that departure to a specific identity. The only way to do that is continuous location surveillance across international checkpoints, travel records, phone signals, and potentially commercial databases. Palantir’s existing platforms have already been used by ICE to analyze driver’s license scans, extract data from seized phones, cross-search location records, and link information from federal, state, and commercial databases into unified investigative files.
ImmigrationOS isn’t a better filing system. It is an AI-coordinated intelligence platform for tracking the movements of human beings at scale.
The Biometric Layer: Clearview and the 60-Billion-Image Database
Running parallel to ImmigrationOS is the facial recognition layer, and its centerpiece is Clearview AI.
Customs and Border Protection holds a $225,000 contract giving federal agents access to Clearview’s database — a corpus of more than 60 billion images scraped from social media, news sites, and the public web. That database has no opt-out mechanism. It was built without consent from any of the people whose faces appear in it. A match against Clearview’s database means your face, taken from a photo you posted years ago, has been used to establish your presence at a place and time.
ICE went further. The agency made a separate $3.75 million contract purchase of Clearview AI technology — the largest single ICE purchase of Clearview’s services to date. The contract buys not just access to the existing database but the capability to run queries against it at volume.
The wider biometric ecosystem extends beyond facial recognition. BI2 Technologies supplies iris-scanning devices used in the field. L3Harris provides stingray technology — devices that impersonate cell towers to capture phone signals and location data from everyone in an area. Paragon Solutions provides phone-hacking software capable of extracting data from locked devices without a password. Each of these tools feeds into the same unified platform architecture that Palantir built.
The Data Broker Pipeline
The surveillance apparatus doesn’t stop at government databases. ICE has acquired tools and relationships that reach into the commercial data ecosystem — the same web of brokers, apps, and platforms that makes your digital footprint visible to advertisers.
PenLink, another vendor in the ICE contractor roster, provides social media monitoring and analysis tools. These systems can track accounts, build connection graphs, and surface location-revealing information from public and semi-public posts. ICE field agents using PenLink can map a target’s social network from a login — and, increasingly, from AI-assisted pattern analysis that doesn’t require a human to review every post.
The license plate reader network adds another layer. As reported in our piece on Home Depot’s Flock Safety cameras, private businesses and local police departments are feeding plate data into shared networks. Those networks connect to fusion centers — regional intelligence hubs that aggregate and redistribute law enforcement data. Flock Safety’s platform, which powers hundreds of thousands of cameras at retailers, apartment complexes, and municipal streets, has documented data-sharing agreements with local agencies. Local agencies, in turn, share with federal partners. The path from your car’s plate to an ICE database is not long, and it doesn’t require a warrant at any step.
Mission Creep: It Was Never Just About Immigration
This is where the story becomes a civil liberties emergency rather than an immigration enforcement debate.
Surveillance tools acquired under an immigration enforcement mandate have already been deployed — and, in at least some cases, were explicitly designed — to monitor U.S. citizens at constitutionally protected activities. Social media monitoring and location tracking systems in the DHS portfolio have been marketed, in the words of publicly available procurement documents, for use “at protests and other gatherings protected under the First Amendment.”
The American Immigration Council has documented specific cases where ICE’s AI tools misidentified individuals — leading to wrongful arrests and months of detention for people who turned out to be citizens or lawful residents. The facial recognition error rate in high-stakes identification contexts remains a documented problem, and Clearview’s database has been independently found to produce incorrect matches at meaningful rates, particularly for people with darker skin tones.
Hundreds of congressional representatives and civil liberties organizations have raised these concerns formally. On April 14, 2026, Representatives Goldman, Wyden, and Velázquez sent a letter to DHS and ICE demanding detailed answers about the scope of the Palantir contracts and which populations the tools are actually being used to monitor. The letter requested, among other things, documentation of how often the tools surface data on U.S. citizens and whether that data is retained.
The agencies have not publicly responded with specifics.
The Constitutional Hook: The Same Question as Chatrie
If you read our April 28 analysis of Chatrie v. United States — the geofence warrant case before the Supreme Court — you’ll recognize the legal architecture underpinning all of this.
The third-party doctrine holds that you have no reasonable expectation of privacy in information you’ve voluntarily shared with a third party. Phone your location to a navigation app, and the government may argue it can access that data without a warrant. Post a photo to Instagram, and the government may argue it can run that photo through a 60-billion-image facial recognition database without constitutional restriction. Hand your license plate to the view of a public camera, and the government may argue the plate is public information — even when the plate is read by an AI camera in a parking lot that feeds into a law enforcement network without your knowledge.
Chatrie will determine whether this logic has limits. If the Supreme Court accepts the government’s broad reading — that opting in to any third-party service eliminates Fourth Amendment protection for the resulting data — then the entirety of ICE’s AI surveillance ecosystem is operating on constitutionally solid ground. The data was “shared.” The users “consented.” The warrant requirement evaporates.
That is the legal theory under which $28.7 billion of surveillance infrastructure has been built. The Supreme Court’s ruling, expected by summer 2026, will determine whether any of it is constitutionally constrained.
What the Scale Actually Means
ICE’s AI-assisted surveillance system is currently targeting more than one million people. Individual contractor companies are receiving up to 50,000 cases per month, with instructions to use “all technology available” to locate each subject before defaulting to traditional investigative methods. That phrase — all technology available — is the operating instruction for a system that spans commercial data brokers, social media scrapers, facial recognition databases, iris scanners, phone-hacking tools, stingray devices, and real-time license plate readers.
The Lieff Cabraser law firm has launched a formal investigation into ICE’s use of Palantir, Clearview AI, Paragon, and PenLink, alleging the combined deployment may constitute illegal data profiling under existing privacy laws. That investigation is early stage. The surveillance infrastructure it’s investigating has been operational for years.
What You Can Do — And What You Can’t
The honest answer is that individual countermeasures are limited against this level of integration. The system isn’t exploiting a single data point. It’s synthesizing dozens of them simultaneously.
That said, some steps remain meaningful. Disabling background location sharing for apps that don’t need it reduces the commercial data broker input. Not using a personally identifiable vehicle in contexts where you want privacy from plate readers is a practical, if inconvenient, approach. Being thoughtful about what photos you post publicly — and where — limits the Clearview input. Encrypted messaging reduces the value of phone-hacking tools if your device is seized.
The more important interventions are political. The Goldman-Wyden-Velázquez letter is the right template — demanding specific disclosure about which populations are being surveilled, under what legal authority, and with what safeguards. State-level legislation restricting commercial data sales to federal agencies has passed or is advancing in several states. Those laws create friction in the data broker pipeline even when federal constitutional protections don’t.
And the Supreme Court’s Chatrie ruling matters here more than almost anywhere else. A ruling that limits the third-party doctrine — or that requires warrants for the kind of aggregated location tracking this system depends on — would force either a legal restructuring of ImmigrationOS or a direct confrontation with the courts over whether warrantless AI surveillance at this scale can stand.
The $28.7 billion has already been spent. The question is whether any of the systems it built will have to comply with the Constitution.
Protect Your Privacy
The surveillance ecosystem described here draws on data you generate every day — location signals, public photos, license plates, phone metadata. Practical steps to reduce your exposure:
- Mobile and location privacy guides — Controls for limiting what apps, carriers, and data brokers collect from your device: MyPrivacy.blog
- Regulatory and compliance frameworks — How federal and state surveillance law works, where restrictions apply, and what rights you have: ComplianceHub.wiki
- Data breach and exposure tracking — Monitoring for when your information surfaces in commercial data broker datasets: Breached.company
For organizations navigating law enforcement data requests, ICE subpoenas, or the intersection of commercial data and federal access, CISO Marketplace provides privacy program assessments, vCISO consulting, and incident response services.
Sources: American Immigration Council reporting on ImmigrationOS and Clearview AI contracts; Immigration Policy Tracking Project contract documentation; Goldman-Wyden-Velázquez letter to DHS, April 14, 2026; Lieff Cabraser investigation announcement; Biometric Update coverage of Congressional scrutiny; WebProNews ICE surveillance budget reporting; The Conversation and Tucson Sentinel analysis of government surveillance expansion; Parriva reporting on ImmigrationOS architecture.
