Algorithmic Targeting in Immigration

How Palantir's ImmigrationOS decides who ICE arrests next.

TL;DR: Immigration and Customs Enforcement (ICE) uses Palantir's "ImmigrationOS" to identify, prioritize, and track people for deportation. The AI system pulls data from 50+ sources, passport records, IRS tax filings, Social Security, medical claims, license plate readers, social media, to build profiles and assign risk scores. There's no public transparency about how the algorithm decides who gets targeted. Former Palantir employees call it unethical. Civil liberties groups warn of embedded bias. And ICE just signed a contract extending it through September 2027.

What Is ImmigrationOS?

Palantir's ImmigrationOS is ICE's next-generation targeting system. It's marketed as a platform to "streamline" deportation decisions. In practice, it's an AI that decides who ICE should go after.[1]

The system does several things:

  • Aggregates data from 50+ sources, Government databases, commercial data brokers, social media, utility records, and location tracking
  • Builds comprehensive profiles, From a single identifier (name, phone number, address), it constructs a complete picture of someone's life
  • Prioritizes targets, Algorithm determines who should be arrested first based on factors ICE won't publicly disclose
  • Tracks voluntary departures, Monitors whether people who agreed to leave actually did
  • Manages deportation workflow, End-to-end case management from identification to removal

ICE awarded Palantir a $30 million contract in April 2025 to develop ImmigrationOS. The prototype was delivered in September 2025. The contract runs through September 2027.[2]

The Data Pipeline

ImmigrationOS doesn't create new surveillance capabilities. It connects existing ones. Here's what feeds into the system:

Government Databases

Passport records, visa applications, border crossing history, IRS tax filings, Social Security records, USCIS immigration files

Commercial Data

Credit reports, utility bills, vehicle registrations, rental applications, address history from data brokers

Location Tracking

License plate readers (Vigilant/Flock), cell phone location from Babel Street, historical movement patterns

Social Media

Public posts, friend networks, check-ins, photos with location data, account creation patterns

The power isn't any single data source. It's the combination. From a phone number, ImmigrationOS can find:[3]

  • Home address (from utility records)
  • Workplace (from location patterns)
  • Family members (from tax records and co-located phones)
  • Daily routine (from license plate reader history)
  • Social network (from social media connections)
  • Immigration status (from USCIS databases)

One identifier becomes a complete life profile in seconds.

The Bias Problem

When an algorithm decides who gets arrested, the algorithm encodes policy. And Palantir won't say what policy ImmigrationOS encodes.

What We Know

  • No public documentation: ICE hasn't published how ImmigrationOS scores or prioritizes individuals
  • No external audits: The system hasn't been reviewed by independent researchers
  • No error rate data: We don't know how often the system makes mistakes
  • No bias testing: No published analysis of whether the system disproportionately targets specific communities

What We Can Infer

The system aggregates data "regardless of its veracity or accuracy."[4] If a database has errors, wrong addresses, mistaken identities, outdated status information, ImmigrationOS treats it as truth.

Machine learning systems inherit bias from their training data. If historical enforcement has disproportionately targeted Latino communities, a system trained on that data will recommend targeting Latinos. This isn't speculation, it's how ML bias works.[5]

The lack of transparency makes this impossible to prove or disprove. That's the point. You can't challenge a black box.

Former Employees Speak Out

In May 2025, thirteen former Palantir employees published an open letter condemning the company's $30 million ICE contract.[6]

Their concerns:

  • The contract "undermines the company's stated principles"
  • AI accelerating immigration enforcement raises "serious ethical concerns"
  • Palantir's neutrality claims are "increasingly difficult to maintain"
  • The platform threatens "democratic norms" beyond immigration

Palantir responded with standard corporate language: they're a "neutral technology provider," clients control their own data, and the company doesn't make enforcement decisions.

CEO Alex Karp has separately defended ICE work, arguing AI tools "help target threats efficiently without broad overreach."[7] He's also made increasingly public statements aligning with Trump administration immigration policy.

Minneapolis: The Algorithm in Action

Operation Metro Surge, the federal deployment that led to Renee Good's death, demonstrates how algorithmic targeting works at scale.[8]

What we know about the Minneapolis targeting:

  • 2,000+ federal agents deployed to Minneapolis-St. Paul
  • ICE described it as their "largest operation ever"
  • Focus on the Somali community specifically
  • Official justification: "fraud" and immigration violations

The surveillance connection:

Mass enforcement operations require target lists. Someone, or something, decides which addresses to raid, which workplaces to sweep, which individuals to prioritize. ImmigrationOS generates those lists.

The Somali community focus suggests algorithmic profiling. When a system is trained to find immigration violators and fed data about ethnic enclaves, it learns to target those enclaves. The algorithm doesn't "see" ethnicity, it sees correlated patterns that amount to the same thing.

We can't prove ImmigrationOS specifically targeted Minneapolis's Somali community. But the infrastructure makes such targeting trivial.

What You Can Do

Minimize Data Footprint

The system aggregates data. Less data = smaller profile. Opt out of data brokers. Disable location tracking on apps. Use cash for sensitive transactions.

Know Your Rights

Algorithmic targeting leads to physical enforcement. Know what to do during ICE encounters. Don't answer questions about immigration status. Don't open doors for I-200 warrants.

Support Transparency Campaigns

Organizations like the ACLU and Immigrant Defense Project have FOIA lawsuits trying to force disclosure of how these systems work. Support them.

Demand Algorithmic Accountability

Push legislators for AI transparency requirements. Some states (California, Colorado) are starting to require algorithmic auditing. Federal law lags behind.

The Bottom Line

ICE doesn't randomly decide who to arrest. An algorithm does.

ImmigrationOS combines 50+ data sources into targeting decisions. It processes information ICE couldn't legally collect directly by buying it from commercial brokers. It makes enforcement "efficient" by prioritizing targets algorithmically. And it operates with zero public transparency about how it decides who gets arrested.

Former Palantir employees call it unethical. Civil liberties groups call it biased. ICE calls it "streamlined enforcement."

The Minneapolis mega-raid that killed Renee Good was powered by this infrastructure. The next raid will be too. Until we demand transparency in algorithmic targeting, we can't challenge the bias embedded in these systems.

The algorithm is already watching. Start making yourself harder to find.

References

  1. American Immigration Council - Palantir ImmigrationOS Analysis (2025)
  2. The Guardian - Palantir Signs $30 Million ICE Contract (April 2025)
  3. Brennan Center - ICE Data Aggregation Capabilities
  4. Cohen Tucker Law - AI in Immigration Enforcement (2025)
  5. The Daily Economy - AI Bias in Immigration (2026)
  6. HR Grapevine - Former Palantir Employees Condemn ICE Contract (May 2025)
  7. Washington Post - Palantir CEO Defends ICE Work (2025)
  8. MPR News - Operation Metro Surge Timeline (January 2026)