Two ICE procurement stories broke within days of each other this month, and read together they describe the current state of American government data purchasing better than any single document.
Story one: the sole-source deal that collapsed
ICE had planned to award Thomson Reuters Special Services a $125 million, five-year sole-source contract — a $25 million base year plus four $25 million options — for investigative analytics built on the CLEAR platform.
CLEAR aggregates public and proprietary records: driver’s licence data, court records, criminal histories, property records, professional licences, utility and address history. ICE’s stated use was a monitoring system providing continuous tracking of up to one million individuals or entities, with event-driven alerts and risk scoring.
That is not a records lookup service. That is a persistent watch capability on a population the size of a large city, assembled from commercial data, with automated notification when a watched person’s circumstances change.
The route to a sole-source award is notable. ICE initially expected to compete the work, received more than fifty industry responses, and identified multiple potentially capable vendors — then rewrote the requirement around proprietary capabilities and concluded that only TRSS could satisfy it. That is the classic shape of a specification written backwards from a chosen vendor.
Then it reversed. ICE said industry feedback prompted it to reconsider before award. A DHS acquisition forecast published 10 August relists the requirement as competitive, values it at more than $100 million, and anticipates a solicitation on 24 August.
The pressure was real: more than two dozen watchdog organisations had called on Congress to investigate the arrangement, and union shareholders had pushed Thomson Reuters directly over CLEAR’s role in deportation operations.
Story two: the deal that did not collapse
While that was happening, ICE moved on a $6.7 million acquisition of LexisNexis records — 82 billion-plus data points from over 10,000 sources.
The requirements language is the story. The database must:
- API-integrate with ICE applications including Palantir, PenLink, and ICE Data Analytics
- support “an artificial intelligence (AI) driven identification system” capable of inferring identity
- enable facial recognition against large-scale image databases in bulk
The stated purpose covers “all aspects of ICE screening and vetting, lead development, and criminal analysis,” including identifying perceived fraud “before crime and fraud can materialize.”
Read that last phrase slowly. The procurement document describes, in its own words, a pre-crime objective.
The downstream destination is the point. Commercial records land in Palantir, where ICE runs ImmigrationOS and, reportedly, ELITE — Enhanced Leads Identification & Targeting for Enforcement — which produces ranked deportation targets with confidence scores. PenLink contributes social media monitoring with neighbourhood-level mobile tracking.
So: a commercial identity graph, an AI inference layer, bulk face matching, social monitoring, and a scoring engine that outputs people to go arrest. ICE arrested over 51,000 people in July 2026.
Why the reversal is not a win
The withdrawal of a sole-source award is a procurement fairness outcome. Competition law exists to ensure the government does not overpay and that vendors get a fair shot. It has nothing to say about whether the capability being bought should exist.
The recompete does not shrink the requirement. The forecast still values it above $100 million and still describes continuous monitoring of up to a million entities. What changes is who gets to build it. TRSS may still win. If it does not, a competitor will, and the competitor’s incentive will be to demonstrate that it can do more for the money.
If anything, competition on a surveillance requirement is a mechanism for capability inflation. Fifty vendors bidding to prove they can track a million people is not a restraining force.
The Fourth Amendment hole this all sits in
None of this requires a warrant, and the reason is the same reason it always is: the government’s position is that purchasing a commercial product is not a search.
Carpenter v. United States held that acquiring historical cell-site location records is a search requiring a warrant, reasoning that the third-party doctrine does not comfortably extend to a comprehensive, effortless, retrospective record of a person’s movements. It did not decide what happens when the government buys equivalent data on the open market rather than compelling it from a carrier.
That gap is the entire business model. Every data-broker contract described above is an argument that Carpenter constrains subpoenas but not invoices. Congress has repeatedly declined to close it — the Fourth Amendment Is Not For Sale Act has passed the House and died in the Senate more than once, and surveillance reform this year has been mostly cosmetic.
And the inputs are getting worse. The DecryptAds launch this week documented how ordinary app advertising feeds the brokers that feed these contracts.
What it means in practice
The pipeline is now explicitly AI-native. Earlier contracts bought records. This one buys records that must integrate with an inference engine and support bulk face matching. The requirement is not “tell us what you know about this person” but “tell us who to look at.”
“Before crime can materialize” is in a government contract. Predictive targeting is no longer a critique applied from outside; it is the stated deliverable. See predictive policing’s track record for how that goes.
Commercial data brokers are federal infrastructure. LexisNexis and Thomson Reuters are not privacy-adjacent companies that happen to have a government line of business. For these purposes they are the sensor layer.
Shareholder and civil-society pressure moved a contract. It did not stop the capability, but it forced a $125 million no-bid award back into the open. That is the first time that lever has visibly worked here.
What you can do
-
Watch the 24 August solicitation. DHS acquisition forecasts and SAM.gov postings are public. The requirements document is the clearest statement of intent any agency ever produces.
-
Opt out of the brokers you can. LexisNexis Risk Solutions and Thomson Reuters both run consumer opt-outs. Both are partial and neither touches the government-facing products, but the consumer-facing records are worth removing.
-
California residents: use DROP. The Delete Request and Opt-out Platform is the only mechanism in the country that hits registered brokers at scale with one request.
-
Cut the supply at the phone. Disable your advertising ID and restrict location permissions. The commercial location layer is fed by devices, not by databases.
-
Ask your representatives about the Fourth Amendment Is Not For Sale Act specifically. Generic privacy advocacy does not move this. The purchase loophole is a named, drafted, repeatedly-shelved fix.
