ICE's New LexisNexis Deal Ties AI Identity Inference to Palantir

Procurement records published on 10 August show that U.S. Immigration and Customs Enforcement (ICE) is paying LexisNexis Risk Solutions $6.7 million for direct, API-level access to a database of “over 82 billion public and proprietary records from more than 10,000 sources” and the ability to push that data into Palantir, PenLink, and ICE Data Analytics. The contract, reported by Joseph Cox at 404 Media, does not just buy records. It buys an “artificial intelligence (AI) driven identification system” that will “infer the identity of the end user” and a facial recognition system that can “perform high accuracy facial matching across diverse sources, including open-source media” and that operates “in bulk.” Federal procurement text is usually hedged; this contract is candid about the surveillance it is buying.

What the contract actually requires

The procurement language is the story. 404 Media obtained the document and quoted the relevant AI and facial recognition clauses directly. Two requirements stand out.

First, the AI identity inference clause. The contract calls for an “artificial intelligence (AI) driven identification system” able to “infer the identity of the end user.” Identity inference is the technical term for resolving a partial identity - a name, an address, a phone number, a face - into a single confident person in the database. What is new is that the procurement text now requires it to be AI-driven rather than analyst-driven, and to be wired into Palantir so that ICE analysts can request identity inferences from inside a Palantir workflow.

Second, the bulk facial matching clause. The procurement text requires facial recognition that uses “large-scale image databases to perform high accuracy facial matching across diverse sources, including open-source media.” “Open-source media” is the unusual phrase. It means social media, news photos, video scraped from the open web - the same broad category of imagery that ICE has separately procured from Clearview AI, whose false matches have already sent innocent people to jail. Per 404 Media, the contract specifies that facial matching must run “in bulk.” Bulk matching is the difference between an officer running a single query against one suspect photo and the agency running continuous matches against every image the database ingests.

The contract is sized for ICE’s Enforcement and Removal Operations (ERO) section: data to support “all aspects of ICE screening and vetting, lead development, and criminal analysis activities.” It names two LexisNexis products - LexID and Accurint Virtual Crime Center - and requires the database to interoperate, by API, with three downstream systems ICE already runs. Per 404 Media, ICE arrested more than 51,000 people in July 2026. LexisNexis, Palantir, and ICE did not immediately respond to 404 Media’s requests for comment.

Why Palantir is the bottleneck

The LexisNexis data is not new to ICE. A May 2026 review of USAspending.gov records by the Berkeley Technology Law Journal found that ICE has held a parallel LexisNexis Risk Solutions contract since 2021, currently running $22.1 million a year, and that ICE searched LexisNexis databases more than 1.2 million times in just seven months of 2021. What is new is the wiring.

The Palantir piece is the integration layer. 404 Media reported in January 2026 that ICE runs a Palantir-built tool called ELITE - “Enhanced Leads Identification & Targeting for Enforcement” - that “populates a map with potential deportation targets,” pulls up a dossier on each person, and assigns a “confidence score” on the person’s current address. The EFF’s January 15, 2026 analysis of the same reporting noted that ELITE “receives peoples’ addresses from the Department of Health and Human Services (which includes Medicaid) and other sources,” and warned about “the Trump administration consolidating all of the government’s information into a single searchable, AI-driven interface with help from Palantir.” EFF Executive Director Cindy Cohn, in an op-ed embedded in the post, called the plan “a throwback to the rightly mocked ‘Total Information Awareness’ plans of the early 2000s.”

The new LexisNexis contract formalizes what the EFF described: it requires the LexisNexis database to talk to Palantir by API. An ICE analyst in Palantir can ask for an identity inference or a bulk facial match and have the response return directly inside the Palantir workflow, alongside the ELITE confidence score on an address and the HHS-derived data already in the dossier. Raw identity data, AI inference, and address confidence become one query path.

What “open-source media” does to the data graph

The phrase “open-source media” is what should worry anyone whose photo is on the internet. The LexisNexis products LexID and Accurint Virtual Crime Center are primarily text-and-record systems: names, addresses, Social Security numbers, license plates, credit headers, employment history, property records. Their facial recognition add-on pulls in faces from “diverse sources, including open-source media” - work-profile photos on LinkedIn, school photos on Facebook, headshots on a company “about” page, any image indexed by an open-web face search engine.

The Berkeley Technology Law Journal review described LexisNexis’s dossier pipeline as compiling records on “every consumer in America.” Bulk facial matching layered on top means a single open-source photo can be resolved against the underlying record, and an analyst in Palantir can request matches continuously rather than one at a time. The 404 Media reporting does not specify a retention period for matched images or a deletion policy when a match turns out to be wrong. LexisNexis Risk Solutions shares a parent with LexisNexis Legal & Professional (the Lexis+ legal-research database) and with Thomson Reuters Special Services; the Berkeley Technology Law Journal noted that “ethical issues arise when lawyers buy and use legal research services sold by the same vendors responsible for building ICE’s surveillance systems” - the same vendors, separate products, same balance sheet.

What This Means

For most readers, the relevant question is whether the database is now wired to the systems that turn identity records into arrest plans. The new contract answers that directly. It requires the LexisNexis database to feed Palantir by API, an AI system that infers identity from partial inputs, and bulk facial matching against images scraped from the open web. Those three requirements are what the EFF in January called a “Total Information Awareness” pattern - a single query path that can resolve an open-source photo into a confident address and feed the result into the tool ICE uses to plan a raid.

For people whose photos are public - which is to say, almost everyone - the practical implication is that the photo itself is now an input to an identity-resolution pipeline. There is no opt-out from the LexisNexis pipeline; the Berkeley Technology Law Journal review notes that the dossier is built from “over 10,000 sources” and includes “every consumer in America.” The new contract adds a face as another input to that pipeline, with the matching done “in bulk.”

For policy, the contract is a concrete procurement record to point at when legislation like Senator Markey’s “ICE Out of Our Faces Act” (S. 3779, introduced 4 February 2026, with Senators Merkley, Wyden, and Jayapal, and companion legislation from Rep. Maxwell Frost in the House) comes up for a vote. The bill would ban ICE and Customs and Border Protection from “acquiring, possessing, accessing, or using any biometric surveillance system, including facial recognition technology.” The procurement text is the proof that bulk facial recognition over open-source media is not hypothetical - it is the explicit requirement on the contract.

The Bottom Line

ICE has bought itself a direct pipe from LexisNexis’s 82-billion-record database into Palantir, with AI identity inference and bulk facial matching across open-source media as the explicit procurement requirements. The contract is $6.7 million. The data is on everyone. The integration is the story.