Lawfare Documents Three DHS AI Systems in Live Immigration Enforcement Without Independent Testing or Disclosure

A Lawfare analysis published August 10, 2026 identified at least three artificial-intelligence systems the Department of Homeland Security is running inside live immigration enforcement, each named and each described as operating without independent testing, public disclosure of its methodology, or external oversight. ImmigrationOS is ICE’s data-fusion targeting platform, which assembles records from across federal and state systems into enforcement targets. The Hurricane Score is an algorithm used in ICE’s alternatives-to-detention program to score individuals; per the analysis it relies on demographic proxies, which is the basis for the equal-protection concern raised. Asylum Text Analytics is a USCIS tool that flags asylum applications and monitors patterns across attorney networks — meaning the system’s surveillance surface extends past the applicant to the lawyers representing them.

Two of the three are not secret in the sense of being undisclosed. DHS’s own published AI use-case inventory lists Hurricane Score as DHS-2408, describing it as modeling the risk that a noncitizen released from detention into the Alternatives to Detention program will fail to comply with check-in requirements — and lists it alongside the Risk Classification Assessment, the automated system ICE has used since 2012 to recommend who is detained and who is released. What the inventory establishes is that the systems exist and what they are for; it does not publish the models, the inputs, the error rates, or any independent validation. The gap Lawfare identifies is between disclosure of existence and disclosure of method.

The precise claim is worth stating narrowly. What Lawfare documents is an absence — no published independent validation, no disclosed error rates, no external review of any of the three — not a finding that a specific statutory or regulatory requirement was violated. That distinction is the argument’s actual point: the systems are deployed in the space where administrative-law oversight would ordinarily attach and does not, because nothing in the current framework requires an agency to test or disclose a model before running it against people. Lawfare frames immigration as the test case precisely because it is where executive AI reaches operational scale first, with the least-protected population and the weakest review.

ImmigrationOS is already in this record as the Palantir-built system tracking U.S. citizens across Medicaid, IRS, and DMV data (2026-04-01–immigrationos-palantir-tracking-us-citizens-medicaid-irs-dmv), procured under the billion-dollar DHS-Palantir surveillance contract (2026-02-19–dhs-signs-billion-dollar-palantir-ice-surveillance-contract) that Senators Wyden, Goldman, and Velázquez had questioned Alex Karp about the prior year (2025-06-17–wyden-goldman-velazquez-letter-karp-palantir-ice-surveillance). The oversight vacuum Lawfare describes was constructed, not inherited: DHS deleted its AI oversight policy in the same month it signed ICE’s largest Clearview AI contract (2026-06-20–ice-largest-clearview-ai-contract-dhs-deletes-oversight-policy). What scoring adds to fusion is a laundering step — an arrest quota enforced through supervisor pressure leaves testimony behind, as it did in Oregon (2025-12-02–ice-agent-testimony-elite-quota-kasubhai-oregon), while the same targeting expressed as a confidence score arrives as a number with no author to depose.